Data audit
We run the same checks on every benchmark: how well we verified its record, who has scored it and how recently, how fresh our read of its leaderboard is, and how far labs' own numbers sit from independent runs. They run on each build, so they follow every sync. Each benchmark's page shows its own audit.
Where are the gaps?
3,311 of 7,002 records pass every check that applies to them. The most common warning is no new score in six months, on 1,716 benchmarks.
| Check | Share of benchmarks | Pass | Warning | Problem | Note |
|---|---|---|---|---|---|
| Catalog record | |||||
| Task access | 6,437 | 565 | |||
| Confidence | 6,448 | 542 | 12 | ||
| Scores | |||||
| Independent runs | 4,777 | 1,310 | 915 | ||
| Models scored | 4,232 | 2,770 | |||
| Newest score | 2,346 | 1,716 | 2,940 | ||
| Frontier models | 674 | 6,328 | |||
| Headroom | 2,175 | 4,827 | |||
| Leaderboard import | |||||
| Importer | 4,766 | 133 | 2,103 | ||
| Last read | 4,762 | 4 | |||
| Read worked | 4,756 | 10 | |||
| Consistency | |||||
| Lab claims | 287 | 26 | 6,689 | ||
| Known issues | 6,995 | 4 | 3 | ||
Every benchmark
Math and science has the largest share of records with a problem, 4 of 415.
- Any status
- Has a problem 25
- Warnings, no problem 3666
- Every check passes 3311
- Any check
- Task access 565
- Confidence 554
- Independent runs 1310
- Newest score 1716
- Importer 133
- Last read 4
- Read worked 10
- Lab claims 26
- Known issues 7
Nothing matches.
- All areas
- Knowledge and language 1455
- Math and science 415
- Coding 739
- Agents and tools 726
- Professional work 1029
- Vision, audio and robotics 1623
- Safety and security 933
- Other 82
| First to look at | |||||
|---|---|---|---|---|---|
| Slop Arena | Knowledge and language | 1 | 3 | 1 of 5 | Confidence Unverified. We haven’t checked it against a primary source yet. |
| Human Pathogen Capabilities Test v2 (HPCT v2) | Safety and security | 1 | 2 | 7 of 10 | Read worked The read on November 18th, 2026 failed with “no chart matches /Model performance on HPCT-v2$/”. Pages show the rows from November 2nd, 2026. |
| Robocurve physical tool-use benchmark (unidentified) | Vision, audio and robotics | 1 | 2 | 1 of 4 | Confidence Unverified. We haven’t checked it against a primary source yet. |
| Brevity Benchmark | Knowledge and language | 1 | 1 | 3 of 5 | Confidence Unverified. We haven’t checked it against a primary source yet. |
| Codeforces rating | Coding | 1 | 1 | 3 of 5 | Confidence Unverified. We haven’t checked it against a primary source yet. |
| DocVQA | Vision, audio and robotics | 1 | 1 | 7 of 9 | Read worked The read on November 18th, 2026 failed with “Command failed: curl --fail --silent --show-error --max-time 40 --user-agent benchmark-almanac (+https://gith…”. Pages show the rows from November 11th, 2026. |
| DUDE | Vision, audio and robotics | 1 | 1 | 8 of 10 | Read worked The read on November 19th, 2026 failed with “Command failed: curl --fail --silent --show-error --max-time 40 --user-agent benchmark-almanac (+https://gith…”. Pages show the rows from November 11th, 2026. |
| GSM8K | Math and science | 1 | 1 | 8 of 10 | Known issues 1 open finding: Contamination. |
| LatchBio Capabilities v1.0 | Math and science | 1 | 1 | 10 of 12 | Confidence Unverified. We haven’t checked it against a primary source yet. |
| PartBench | Professional work | 1 | 1 | 3 of 5 | Confidence Unverified. We haven’t checked it against a primary source yet. |
| Roboharm | Vision, audio and robotics | 1 | 1 | 7 of 9 | Confidence Unverified. We haven’t checked it against a primary source yet. |
| StrataBench Land v1 Extract | Professional work | 1 | 1 | 7 of 9 | Read worked The read on November 19th, 2026 failed with “401 https://huggingface.co/datasets/Collide-ai/stratabench-land-v1/raw/main/README.md”. Pages show the rows from November 8th, 2026. |
| ARC-AGI-3 | Agents and tools | 1 | 9 of 10 | Known issues 1 open finding: Harness differences and unclear tasks. | |
| CathedralBench | Safety and security | 1 | 4 of 5 | Confidence Unverified. We haven’t checked it against a primary source yet. | |
| Clocktower Radio | Agents and tools | 1 | 8 of 9 | Read worked The read on November 19th, 2026 failed with “403 https://clocktower-radio.com/”. Pages show the rows from November 18th, 2026. | |
| ExploitBench | Safety and security | 1 | 9 of 10 | Read worked The read on November 19th, 2026 failed with “403 https://exploitbench.ai/”. Pages show the rows from November 18th, 2026. | |
| HLE-Diamond | Knowledge and language | 1 | 11 of 12 | Known issues 1 open finding: Leaked answers and reward hacking. | |
| IOI 2025 | Coding | 1 | 3 of 4 | Confidence Unverified. We haven’t checked it against a primary source yet. | |
| OpenAI Deployment Simulation | Safety and security | 1 | 6 of 7 | Confidence Unverified. We haven’t checked it against a primary source yet. | |
| SciReasoner evaluation suite | Math and science | 1 | 4 of 5 | Confidence Unverified. We haven’t checked it against a primary source yet. | |
| Story Cloze Winter 2018 | Knowledge and language | 1 | 7 of 8 | Read worked The read on November 19th, 2026 failed with “fetch failed”. Pages show the rows from November 18th, 2026. | |
| StoryCloze | Knowledge and language | 1 | 7 of 8 | Read worked The read on November 19th, 2026 failed with “fetch failed”. Pages show the rows from November 18th, 2026. | |
| StudyArena Text | Knowledge and language | 1 | 9 of 10 | Read worked The read on November 20th, 2026 failed with “500 https://studyarena.com/leaderboard”. Pages show the rows from November 18th, 2026. | |
| SuperMatBench | Math and science | 1 | 4 of 5 | Confidence Unverified. We haven’t checked it against a primary source yet. | |
| XTREME | Knowledge and language | 1 | 7 of 8 | Read worked The read on November 18th, 2026 failed with “502 https://sites.research.google/xtreme/app.js”. Pages show the rows from November 18th, 2026. | |
| Adaptive Reasoning (Vellum) | Knowledge and language | 3 | 2 of 5 | Importer The record links a board on vellum.ai that we don’t read yet. | |
| AL-QASIDA | Knowledge and language | 3 | 2 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Alpha Arena | Professional work | 3 | 2 of 5 | Importer The record links a board on alphaarena.ai that we don’t read yet. | |
| AlphaEval | Professional work | 3 | 7 of 10 | Task access We don’t know whether the tasks are public. | |
| Artificial Analysis Image Arena | Vision, audio and robotics | 3 | 2 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| Artificial Analysis Progress Index | Other | 3 | 1 of 4 | Task access We don’t know whether the tasks are public. | |
| AudioBench | Vision, audio and robotics | 3 | 2 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| BigLaw Bench: Arena | Professional work | 3 | 2 of 5 | Importer The record links a board on harvey.ai that we don’t read yet. | |
| CC-Bench | Coding | 3 | 6 of 9 | Task access We don’t know whether the tasks are public. | |
| Cozzo AI Receptionist Benchmark (Round 4) | Agents and tools | 3 | 2 of 5 | Importer The record links a board on cozzo.io that we don’t read yet. | |
| DeepResearch Arena | Agents and tools | 3 | 6 of 9 | Task access We don’t know whether the tasks are public. | |
| Dynabench SQuAD | Knowledge and language | 3 | 2 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Fiction.LiveBench | Knowledge and language | 3 | 3 of 6 | Importer The record links a board on fiction.live that we don’t read yet. | |
| Fireworks Specialized Intelligence Index | Other | 3 | 2 of 5 | Importer The record links a board on fireworks.ai that we don’t read yet. | |
| IChO 2025 | Math and science | 3 | 2 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Jigsaw Toxic Comment Classification | Safety and security | 3 | 3 of 6 | Importer The record links a board on kaggle.com that we don’t read yet. | |
| LARYBench | Knowledge and language | 3 | 1 of 4 | Task access We don’t know whether the tasks are public. | |
| MedBench 4.0 (OpenCompass) | Professional work | 3 | 1 of 4 | Task access We don’t know whether the tasks are public. | |
| METR general autonomy task suite | Coding | 3 | 2 of 5 | Task access We don’t know whether the tasks are public. | |
| NoisyBench | Knowledge and language | 3 | 7 of 10 | Task access We don’t know whether the tasks are public. | |
| Polaris Healthcare Safety Benchmarks | Professional work | 3 | 2 of 5 | Importer The record links a board on hippocraticai.com that we don’t read yet. | |
| STALE | Knowledge and language | 3 | 7 of 10 | Task access We don’t know whether the tasks are public. | |
| Stick to Your Role! | Knowledge and language | 3 | 6 of 9 | Task access We don’t know whether the tasks are public. | |
| Subversion Strategy Eval | Safety and security | 3 | 2 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| SycophancyEval (Sharma et al.) | Safety and security | 3 | 2 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Translated ARC-Challenge, HellaSwag and MMLU (Mistral) | Knowledge and language | 3 | 2 of 5 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: pingdotgg/t3code | Coding | 3 | 2 of 5 | Importer An importer exists, but it hasn’t read the board yet. | |
| WebQuest | Vision, audio and robotics | 3 | 2 of 5 | Task access We don’t know whether the tasks are public. | |
| WorldModelGym | Vision, audio and robotics | 3 | 2 of 5 | Importer The record links a board on wm-gym.labs.reka.ai that we don’t read yet. | |
| 2018 n2c2 Cohort Selection for Clinical Trials | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| 2024 NIST GenAI Text-to-Text Generators | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on miganalytics.nist.gov that we don’t read yet. | |
| 34th Japanese National Examination for Dental Hygienists | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| 3D Animation Arena | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| 3DSRBench | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| A-Bench | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| A.S.E 2.0 | Safety and security | 2 | 3 of 5 | Importer The record links a board on aicgseceval.tencent.com that we don’t read yet. | |
| A^3-Bench | Math and science | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| AA-Image-Editing | Vision, audio and robotics | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| AA-Video-Editing | Vision, audio and robotics | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Act2Answer | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| AD4Edu-Bench | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| AdaPlanBench | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Advanced Sommelier (theory knowledge) | Professional work | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| AdvBench | Safety and security | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| Adversarial SQuAD | Knowledge and language | 2 | 3 of 5 | Independent runs No scores from any source yet. | |
| Aether-Bench | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| AgentClinic | Professional work | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| AgentCLUE | Agents and tools | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| AgentCLUE-General | Agents and tools | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| AgentCLUE-Mobile (offline) | Agents and tools | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| AgenticDataBench | Coding | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| AgentKernelArena | Coding | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| AGI Score (A Definition of AGI) | Other | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| AGORA (Agentic Global Open Research Arena) | Professional work | 2 | 3 of 5 | Importer The record links a board on agoraresearcharena.org that we don’t read yet. | |
| Agronomy AI Arena | Professional work | 2 | 3 of 5 | Importer The record links a board on agronomyarena.com that we don’t read yet. | |
| AI AgriBench Smallholders | Professional work | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| AI Diplomacy | Agents and tools | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| AI-Spire (AGI-Eval) | Agents and tools | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| AI4D Malawi News Classification Challenge | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on zindi.world that we don’t read yet. | |
| AidanBench | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| AIME 2023 | Math and science | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| AInsteinBench | Coding | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| AIR-Bench 24.04 | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| Alignment faking evaluation, helpful-only setting (Anthropic) | Safety and security | 2 | 6 of 8 | Task access We don’t know whether the tasks are public. | |
| ALIVE-Bench | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| All-Angles Bench | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Almieyar | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| AMC 12 2024 | Math and science | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| AMUSE | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Android in the Zoo (AitZ) | Agents and tools | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Anthropic cyber safeguard evaluations | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Anthropic harmful-request and wellbeing evaluations | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Anthropic internal life-sciences evals | Math and science | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Anthropic large agent teams tasks | Agents and tools | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Anthropic malicious agentic use evaluations | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Anthropic targeted alignment and honesty evaluations | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| AP Art History | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP Biology | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP Calculus | Knowledge and language | 2 | 1 of 3 | Task access We don’t know whether the tasks are public. | |
| AP Calculus BC | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP Chemistry | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP English Language and Composition | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP English Literature and Composition | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP Environmental Science | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP Macroeconomics | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP Microeconomics | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP Physics | Knowledge and language | 2 | 1 of 3 | Task access We don’t know whether the tasks are public. | |
| AP Physics 2 | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP Psychology | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP Statistics | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP US Government | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP US History | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AP World History | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Apodex Discovery | Math and science | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| Apollo Research coding agent monitor calibration | Safety and security | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AQuA | Coding | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AraDiCE | Knowledge and language | 2 | 3 of 5 | Independent runs No scores from any source yet. | |
| ArAIEval 2023 | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on docs.google.com that we don’t read yet. | |
| ArAIEval 2024 | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on docs.google.com that we don’t read yet. | |
| Artificial Analysis Intelligence Index v1 | Other | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Artificial Analysis Intelligence Index v2.2 | Other | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Artificial Analysis Intelligence Index v4.2 | Other | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Artificial Analysis Speech Arena | Vision, audio and robotics | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Artificial Analysis Video Arena | Vision, audio and robotics | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| ArXivPhys | Math and science | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Aspire | Coding | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| AssertionForge design evaluation | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| AudioMOS Challenge 2025 | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on docs.google.com that we don’t read yet. | |
| AutoBench Agronomy LLM Benchmark | Professional work | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| AutoCodeBench | Coding | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Automated LLM Speedrunning Benchmark | Coding | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| AutoPenBench | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| AutoResearchBench | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| AV-SafetyBench | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| AVMeme Exam | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| BABE | Safety and security | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| BABILong | Knowledge and language | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Bamboogle | Knowledge and language | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| BashBench | Safety and security | 2 | 3 of 5 | Importer The record links a board on bashcontrol.com that we don’t read yet. | |
| BIBLE Translation Benchmark (Meta) | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| BIG-bench | Knowledge and language | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| BigLaw Bench Retrieval | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| BigLaw Bench: Global | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| BigLaw Bench: Research | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| BioASQ | Professional work | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| BiomedArena | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| BioRiskEval | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| BrowserART | Safety and security | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Bug in the Code Stack | Knowledge and language | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| BuzzBench | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| CAPBENCH-English | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on uspolicyagendas.com that we don’t read yet. | |
| CapQuiz | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| CareManage-Bench | Professional work | 2 | 3 of 5 | Importer The record links a board on kaggle.com that we don’t read yet. | |
| CascadeBench | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| CEFR Adherence Benchmark | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on kaggle.com that we don’t read yet. | |
| Cekura Bench | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on benchmarks.cekura.ai that we don’t read yet. | |
| CELLO | Knowledge and language | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| CG-Bench | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| Chain-of-thought faithfulness with clues (Anthropic) | Safety and security | 2 | 6 of 8 | Task access We don’t know whether the tasks are public. | |
| ChainWorld | Agents and tools | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| ChartMimic | Coding | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Chartography | Vision, audio and robotics | 2 | 9 of 11 | Lab claims 7 of 10 pairs differ by more than 10 points, the largest for Claude Opus 5 (−55 pts). | |
| ChartX | Vision, audio and robotics | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| ChemCoTBench | Math and science | 2 | 3 of 5 | Importer The record links a board on howardli1984.github.io that we don’t read yet. | |
| ChemInsight | Math and science | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| ChemPro | Math and science | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| Chest ImaGenome | Professional work | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| CHID | Knowledge and language | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| ChJGBench | Professional work | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| CIPHERGRID | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Civil-Bench | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| CLBench-V | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| CLEAR (Clinical Information Extraction) | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| CLEF HIPE 2022 | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on hipe-eval.github.io that we don’t read yet. | |
| CLEF JOKER 2026 Task 4: Humour Generation | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| CMMMU | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| CMO 2024 | Math and science | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| CMO 2025 | Math and science | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| CNMO 2024 | Math and science | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| CoCoNot | Safety and security | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| CodeSimpleQA | Knowledge and language | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| CompactionBench | Agents and tools | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| COMPASS | Agents and tools | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| ComplexNLU (Tencent) | Knowledge and language | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Contextual Invoice OCR | Professional work | 2 | 3 of 5 | Importer The record links a board on builders.ramp.com that we don’t read yet. | |
| ContextualBench | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| ContinualSkillBench | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Contract Intelligence | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Copilot Arena | Coding | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Copilot Arena Code Editing | Coding | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| Cost Per Correct | Agents and tools | 2 | 3 of 5 | Importer The record links a board on kaggle.com that we don’t read yet. | |
| CounselBench-100 | Professional work | 2 | 3 of 5 | Importer The record links a board on blobfish.ai that we don’t read yet. | |
| Coval Voice AI Benchmarks | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on benchmarks.coval.ai that we don’t read yet. | |
| CRAB | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| CRAG | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Creation-MMBench | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| CRPE | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| CTIBench | Safety and security | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| CutVerse | Agents and tools | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| CV3-Eval | Vision, audio and robotics | 2 | 3 of 5 | Independent runs No scores from any source yet. | |
| CVQA | Knowledge and language | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| CVTG-2K | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| CyberSecEval Autonomous Offensive Cyber Operations | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| CyberSecEval MITRE | Safety and security | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| CyberSecEval MITRE False Refusal Rate | Safety and security | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| CyberSecEval Visual Prompt Injection | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| CyberSecEval Vulnerability Exploitation | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| DAComp | Coding | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| DailyReport | Agents and tools | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| data-eng-bench | Coding | 2 | 3 of 5 | Importer The record links a board on hub.harborframework.com that we don’t read yet. | |
| DataCLUE | Coding | 2 | 3 of 5 | Importer The record links a board on github.com that we don’t read yet. | |
| DAXBench v2 | Coding | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| DEBAGREEMENT | Knowledge and language | 2 | 3 of 5 | Importer An importer exists, but it hasn’t read the board yet. | |
| DecepEval | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Deceptive Reasoning Exposure Suite | Safety and security | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| DecodingTrust | Safety and security | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| DeepAmbigQA | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| DeepSeek Chinese creative writing evaluation | Knowledge and language | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| DELEGATE-52 | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Dementia Family Caregiver Detection | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| DeR2 Bench | Agents and tools | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| DesignArena Speech-to-Speech | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on designarena.ai that we don’t read yet. | |
| DETOUR | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| DevOpsBench-100 | Agents and tools | 2 | 3 of 5 | Importer The record links a board on blobfish.ai that we don’t read yet. | |
| DI-Bench | Professional work | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| DiscoX | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| Discrim-Eval | Safety and security | 2 | 3 of 5 | Independent runs No scores from any source yet. | |
| Disengagement-Aware Student Simulators | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| DISPLACE Challenge 2023 | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on codalab.lisn.upsaclay.fr that we don’t read yet. | |
| Domain Intelligence Benchmark Suite | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| DPG-Bench | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Drawing with LLMs | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on kaggle.com that we don’t read yet. | |
| Dual Exam | Vision, audio and robotics | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| DuMateBench | Professional work | 2 | 3 of 5 | Importer The record links a board on dumatebench.com that we don’t read yet. | |
| Dutch DPA national voting advice audit | Safety and security | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Dynamic-SUPERB Phase-2 | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| EEBench | Professional work | 2 | 10 of 12 | Task access We don’t know whether the tasks are public. | |
| EgoProceVQA | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| EIFBench | Knowledge and language | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| EmailBench | Professional work | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| EmbodiedCLUE-VLA (tabletop tasks) | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| Encyclo-K | Knowledge and language | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| Enterprise PII | Safety and security | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| Enterprise Scenarios Creative Writing | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| Enterprise Scenarios Customer Support Dialogue | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| Enterprise Scenarios Leaderboard | Professional work | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| Enterprise Scenarios Toxicity | Safety and security | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| Enterprise Task Pareto Frontier (Glean) | Professional work | 2 | 3 of 5 | Importer The record links a board on glean.com that we don’t read yet. | |
| EnvSimBench | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| ERR Estonian language and cultural history questions | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| ESI-Bench | Vision, audio and robotics | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| ETVA | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| euchre-bench | Coding | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| EvaHan 2023 | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Evidence Sufficiency Evaluation | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on kaggle.com that we don’t read yet. | |
| EvoArena | Knowledge and language | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| ExeWorldBench | Vision, audio and robotics | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| EXP-Bench | Coding | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| ExpertEval | Professional work | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| FAD Music Listening Test | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| FaithJudge | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| FieldWorkArena V2.0 | Professional work | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| FilBench | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| Finance Fundamentals | Professional work | 2 | 3 of 5 | Importer The record links a board on benchmarks.kensho.com that we don’t read yet. | |
| Financial Statement OCR | Professional work | 2 | 3 of 5 | Importer The record links a board on builders.ramp.com that we don’t read yet. | |
| FinDeepIndicator | Professional work | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| FinEval | Professional work | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| FlashInfer-Bench | Coding | 2 | 6 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| FormExtract Bench | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on kaggle.com that we don’t read yet. | |
| FORTRESS-RN | Safety and security | 2 | 3 of 5 | Importer The record links a board on labs.scale.com that we don’t read yet. | |
| FrontierMath: Open Problems | Math and science | 2 | 3 of 5 | Importer The record links a board on epoch.ai that we don’t read yet. | |
| FutureEval (Metaculus) | Math and science | 2 | 3 of 5 | Importer The record links a board on metaculus.com that we don’t read yet. | |
| GateMem | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Gemini agentic alignment evals | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Gemini content safety evals | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Gemini Personality Evaluation (sycophancy) | Safety and security | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Gemini prompt-injection evaluation suites | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| GenAI-Bench | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| GeneBench | Safety and security | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| Generating Confusion: Voting with a Disability | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on cdt.org that we don’t read yet. | |
| GeoFlip | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| German federal election chatbot audit | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Google DeepMind harmful manipulation studies | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Google DeepMind internal cyber benchmarks | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Gray Swan Shade (adaptive attacker) | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| GRE Analytical Writing | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| GRE Physics | Math and science | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| GRE Quantitative | Math and science | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| GRE Verbal | Knowledge and language | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| GroundUI | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| GSR-Bench | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| HalluWorld | Other | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| Harness Leaderboard | Agents and tools | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| HarnessDev | Coding | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| HCAST | Coding | 2 | 3 of 5 | Importer The record links a board on metr.org that we don’t read yet. | |
| HealthBench Hard Hallucinations | Knowledge and language | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| HEART | Safety and security | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| HELM Arabic Enterprise | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| Hermes Bench | Agents and tools | 2 | 9 of 11 | Task access We don’t know whether the tasks are public. | |
| Hindsight Neglect | Knowledge and language | 2 | 3 of 5 | Independent runs No scores from any source yet. | |
| HMMT February 2024 | Math and science | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| HubBench | Professional work | 2 | 3 of 5 | Importer The record links a board on blobfish.ai that we don’t read yet. | |
| Hugging Face Speech Bench | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| Human Pathogen Capabilities Test | Safety and security | 2 | 10 of 12 | Task access We don’t know whether the tasks are public. | |
| HumanEval-COBOL (Cohere) | Coding | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| HumanEval-V | Coding | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Husky Hold'em Bench | Agents and tools | 2 | 3 of 5 | Importer The record links a board on huskybench.com that we don’t read yet. | |
| ICDAR 2025 MapText Competition | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on rrc.cvc.uab.es that we don’t read yet. | |
| ICPC World Finals 2025 | Coding | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| IFEval-OOD | Knowledge and language | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| ImageMining | Agents and tools | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| IMO Shortlist 2024 | Math and science | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| InfiHall-Bench | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| IntelligentBench | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| Internal API instruction following (OpenAI) | Knowledge and language | 2 | 9 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| IrokoBench | Knowledge and language | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Japanese Nephrology Board Renewal Questions 2014 to 2023 | Professional work | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| Joi-Bench-Task | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| K-BrowseComp | Knowledge and language | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| KellyBench | Math and science | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Kili LLM Reasoning Benchmark | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Kimi Chinese in-house open-ended benchmark | Knowledge and language | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Kimi-Audio speech conversation evaluation | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| KMMLU-Hard | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Knowledge Cutoff Benchmark | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on apoorvumang.github.io that we don’t read yet. | |
| Ko-Arena-Hard v2 (Upstage) | Knowledge and language | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Kontext Bench | Vision, audio and robotics | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Kor-Phishing | Safety and security | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Korean Canonical Legal Benchmark | Professional work | 2 | 3 of 5 | Importer An importer exists, but it hasn’t read the board yet. | |
| Korean Dental Licensing Examination 2024 and 2025 | Professional work | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| Korean National Licensing Examination for Korean Medicine Doctors 2022 | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| LAB: Law Firm Knowledge | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| LAB: US Case Law | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| LAD2000 | Vision, audio and robotics | 2 | 3 of 5 | Importer An importer exists, but it hasn’t read the board yet. | |
| LedgerBench-100 | Professional work | 2 | 3 of 5 | Importer The record links a board on blobfish.ai that we don’t read yet. | |
| LeetCode Weekly Contest | Coding | 2 | 9 of 11 | Task access We don’t know whether the tasks are public. | |
| LeWitt-Bench | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on designarena.ai that we don’t read yet. | |
| LHAW | Agents and tools | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| LibriSpeech | Vision, audio and robotics | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| LitBench | Knowledge and language | 2 | 3 of 5 | Importer An importer exists, but it hasn’t read the board yet. | |
| LiveMACE | Professional work | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| LiveResearchBench | Agents and tools | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| LiveSecBench | Safety and security | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| LLM Colosseum | Agents and tools | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| LLM Economist | Professional work | 2 | 3 of 5 | Importer An importer exists, but it hasn’t read the board yet. | |
| LLMBoxing | Knowledge and language | 2 | 6 of 8 | Task access We don’t know whether the tasks are public. | |
| Long-document QA (Kensho) | Professional work | 2 | 3 of 5 | Importer The record links a board on benchmarks.kensho.com that we don’t read yet. | |
| LongBench-Chat | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| LongMemEval-V2 | Knowledge and language | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| LongPuzzleBench | Agents and tools | 2 | 3 of 5 | Importer An importer exists, but it hasn’t read the board yet. | |
| LongText-Bench | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| LPFQA | Knowledge and language | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| LSAT | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| LVBench (with memory) | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on lvbench.github.io that we don’t read yet. | |
| Mandarin-Minority translation (Tencent) | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| ManiLadder | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on maniladder-benchmark.github.io that we don’t read yet. | |
| Mantic vs Polymarket backtest | Math and science | 2 | 3 of 5 | Importer The record links a board on blog.mantic.com that we don’t read yet. | |
| Matbench | Math and science | 2 | 3 of 5 | Importer The record links a board on matbench.materialsproject.org that we don’t read yet. | |
| Math401 | Math and science | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| MathCanvas-Bench | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| MathCLUE | Math and science | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| MC-Bench | Vision, audio and robotics | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| MedBench (Russian closed-test platform) | Professional work | 2 | 3 of 5 | Importer The record links a board on medbench.ru that we don’t read yet. | |
| MedBench 2.0 (OpenCompass) | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| MedBench 3.0 (OpenCompass) | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| MedBench 5.0 (OpenCompass) | Professional work | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| Medical Knowledge Self-Assessment Program (MKSAP) | Professional work | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| MemEye | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| MemLens | Knowledge and language | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| MEMTRACK | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| MENT-SNS | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| METR data deduplication | Coding | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| METR pre-deployment AI R&D tasks | Coding | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| METR side task monitorability evaluation | Safety and security | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| micro1 Medical Index | Professional work | 2 | 9 of 11 | Task access We don’t know whether the tasks are public. | |
| MIMIC-III Report Summarization | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| MIMIC-IV-BHC | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| MiMo Coding Bench | Coding | 2 | 9 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| MindEval | Professional work | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| MindEval2 | Professional work | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| MINTEval | Knowledge and language | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| MirageBench | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| MM-HELIX | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| MM-SafetyBench | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| MMBench | Vision, audio and robotics | 2 | 7 of 9 | Lab claims 1 of 4 pairs differs by more than 10 points, the largest for Janus-Pro-7B (+17 pts). | |
| MMBench-GUI | Agents and tools | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| MMBench-Video | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| MME-CC | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| MME-CoT | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| MMSearch | Agents and tools | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| MobileMiniWoB++ | Agents and tools | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Model Slant | Safety and security | 2 | 3 of 5 | Importer The record links a board on modelslant.com that we don’t read yet. | |
| Moneda Bench | Professional work | 2 | 3 of 5 | Importer The record links a board on benchmarks.finance that we don’t read yet. | |
| MORSE-500 | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| Movie Gen Video Bench | Vision, audio and robotics | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| MR-NIAH (Multi-Round Needles-In-A-Haystack) | Knowledge and language | 2 | 3 of 5 | Independent runs No scores from any source yet. | |
| MSRVTT-QA | Vision, audio and robotics | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Multi-LCB | Coding | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| MultiCardioNER | Professional work | 2 | 3 of 5 | Importer The record links a board on participants-area.bioasq.org that we don’t read yet. | |
| Multilingual MMBench | Knowledge and language | 2 | 3 of 5 | Independent runs No scores from any source yet. | |
| MuseBench (Music Score and Performance Understanding) | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| MVDriveBench | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| NanoGPT Expenditure Horizon | Coding | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| NaturalBench | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| NaturalCodeBench | Coding | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| NeedleBench | Knowledge and language | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Nejumi LLM Leaderboard 4 | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on wandb.ai that we don’t read yet. | |
| NIA Sovereign AI Benchmark Evaluation, Phase 1 | Other | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| NLPCC 2023 Chinese Essay Discourse Coherence Evaluation | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on github.com that we don’t read yet. | |
| Nuscene | Vision, audio and robotics | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| O-NET M6 | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on openthaigpt.aieat.or.th that we don’t read yet. | |
| OBLIQ-Bench | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| ODV-Bench | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| OlympicArena | Math and science | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| OmniMatBench | Math and science | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| OmniMMI | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| OmniRouting | Professional work | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| OncoTriad-QA | Professional work | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| OneIG-Bench | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Open LLM Leaderboard v2 | Other | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| OpenAI agentic alignment evals | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| OpenAI cyber safety evaluations (incl. Daybreak Blue) | Safety and security | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| OpenAI Dynamic Mental Health Benchmarks | Safety and security | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| OpenAI GPT-Red jailbreak and prompt-injection evals | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| OpenAI internal AI self-improvement evals | Coding | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| OpenAI internal cyber evaluations | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| OpenAI Production Benchmarks and safe-completion evals | Safety and security | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| OpenComputer | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| OpenScore String Quartets | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Paloma | Knowledge and language | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Pare-Bench | Agents and tools | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Partner Restrictions Compliance | Professional work | 2 | 3 of 5 | Importer The record links a board on builders.ramp.com that we don’t read yet. | |
| PAWBench | Vision, audio and robotics | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| PAWS | Knowledge and language | 2 | 3 of 5 | Independent runs No scores from any source yet. | |
| PerMemBench | Knowledge and language | 2 | 6 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| Persian MT-Bench | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| PhAIL | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| PharmaBench (micro1) | Professional work | 2 | 9 of 11 | Task access We don’t know whether the tasks are public. | |
| PhysCapBench | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on hfwang2001.github.io that we don’t read yet. | |
| PhysicalRealismBench-U | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| PhyWorldBench | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| PlanBench-XL | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| PlanningBench | Knowledge and language | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Progress-Bench | Vision, audio and robotics | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| ProtocolQA Open-Ended (xAI) | Safety and security | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| PROWBench | Vision, audio and robotics | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| PUMaC 2024 | Math and science | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Q-Bench | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Q-Bench-Video | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| QF-Bench | Professional work | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Questflow Traders Arena | Professional work | 2 | 3 of 5 | Importer The record links a board on next.questflow.ai that we don’t read yet. | |
| Qwen-Audio-Agent In-House Cockpit Benchmark | Agents and tools | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Qwen-Music Text-to-Music Generation | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Qwen2 multilingual safety evaluation | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| ReactHuman | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| RealmPathology | Professional work | 2 | 9 of 11 | Task access We don’t know whether the tasks are public. | |
| ReasoningJudgeBench | Coding | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Red-Teaming Resistance Benchmark | Safety and security | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| RefineBench | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| RelBench | Coding | 2 | 4 of 6 | Importer The record links a board on star-project.stanford.edu that we don’t read yet. | |
| RepliBench | Coding | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| RepoChat Arena | Coding | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| ReTrace | Agents and tools | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| Reward-seeking via contrastive belief updates | Safety and security | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| RH SWE-bench | Coding | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| RiOSWorld | Safety and security | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| RoboEval | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| RoboMemArena | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Role-play Benchmark (MiniMax) | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Roman Urdu Code-Switch Benchmark | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on kaggle.com that we don’t read yet. | |
| RTP-LX | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| RTVLM | Safety and security | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| S3Gym | Coding | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| SA-1B Hard human pose evaluation | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| SafeLIBERO | Vision, audio and robotics | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| SAM 3D Artist Objects | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| SAM Audio Judge evaluation | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| SAT Evidence-Based Reading & Writing | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| SAT Math | Math and science | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| SAT Reading | Knowledge and language | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| SatLidar canopy height evaluation | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| SciPredict | Math and science | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SciTS | Math and science | 2 | 3 of 5 | Independent runs No scores from any source yet. | |
| SCONE-bench | Safety and security | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| SecureBio VMQA, single choice | Safety and security | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| SEED-Bench-2 | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SEED-Bench-2-Plus | Vision, audio and robotics | 2 | 9 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| SemComp-Bench | Vision, audio and robotics | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| SFI-Bench | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Single-turn election integrity (Anthropic, 2026) | Safety and security | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| SITE | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SLATE | Professional work | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| SMM4H 2025 Food Recalls and Foodborne Disease Outbreaks | Professional work | 2 | 6 of 8 | Task access We don’t know whether the tasks are public. | |
| SnorkelGraph | Math and science | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SnorkelSequences | Math and science | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SOL-ExecBench | Coding | 2 | 3 of 5 | Importer The record links a board on research.nvidia.com that we don’t read yet. | |
| Sonetra Norwegian SNOMED CT Term Recognition | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| SORRY-Bench | Safety and security | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| SpaceXAI internal R&D evals | Coding | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| SpatialEval Maze-Nav | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SpatialEval Spatial-Map | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SpatialWorld | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SpeechRole | Vision, audio and robotics | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| SpeedrunBench | Agents and tools | 2 | 3 of 5 | Importer The record links a board on patronus.ai that we don’t read yet. | |
| Spider2-V | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SpireBench | Agents and tools | 2 | 3 of 5 | Importer An importer exists, but it hasn’t read the board yet. | |
| SPM-Bench | Math and science | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| States United voter questions | Knowledge and language | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Strategy Bench | Agents and tools | 2 | 3 of 5 | Importer The record links a board on strategy.freysa.ai that we don’t read yet. | |
| StreamBench (StreamChat) | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| StreamEval | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on streameval.ai that we don’t read yet. | |
| SuperCLUE-200K | Knowledge and language | 2 | 6 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Auto | Professional work | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Code3 | Coding | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-CoT | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-CPIF | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-DeepSearch | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-DSPSafeBench | Safety and security | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Fact | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Faith | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Hard | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-ICabin | Agents and tools | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Industry | Professional work | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Long | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Math6 | Math and science | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| SuperCLUE-Math6o | Math and science | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Mkt | Professional work | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Open | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-OpenClaw | Agents and tools | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Project | Coding | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Role | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Science | Math and science | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-V | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Video | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-VLR | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Voice (speech interaction) | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-WebCoding | Coding | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Writing | Knowledge and language | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| SVAMP | Math and science | 2 | 3 of 5 | Importer An importer exists, but it hasn’t read the board yet. | |
| SWE-EVO | Coding | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| SWE-Explore | Coding | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SWE-Interact | Coding | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SwitzerlandQA | Knowledge and language | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| SycophancyEval (answer sycophancy) | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| SynthDocBench | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Talking Points: When Chatbots Surface Russian State Media | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on isdglobal.org that we don’t read yet. | |
| TemporalBench | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| TERMS-Bench | Agents and tools | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| TextArena | Agents and tools | 2 | 3 of 5 | Importer The record links a board on textarena.ai that we don’t read yet. | |
| TextVBench | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| The Midnight Index | Safety and security | 2 | 3 of 5 | Importer The record links a board on secondstrike.io that we don’t read yet. | |
| The Turtle Test | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on kaggle.com that we don’t read yet. | |
| TIIF-Bench | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Toloka Arena | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| TrafficImag | Vision, audio and robotics | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| TRAIT | Safety and security | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| ts-bench | Coding | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| TSP-Opt | Math and science | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| TTS Arena | Vision, audio and robotics | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| TutorBench | Other | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Uhura ARC-Easy | Knowledge and language | 2 | 3 of 5 | Independent runs No scores from any source yet. | |
| Uhura TruthfulQA | Knowledge and language | 2 | 3 of 5 | Independent runs No scores from any source yet. | |
| Uhura-Eval | Knowledge and language | 2 | 3 of 5 | Independent runs No scores from any source yet. | |
| UI-Genie reward model evaluation | Agents and tools | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| UK AISI alignment and monitorability evaluations | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| UltraText Bench | Vision, audio and robotics | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Uniform Bar Exam (MBE+MEE+MPT) | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| USABO Semifinal Exam 2020 | Math and science | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| USAMO 2024 | Math and science | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| USNCO Local Section Exam 2022 | Math and science | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Vaani Benchmark V1.0 | Vision, audio and robotics | 2 | 3 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| VAKRA | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Vals Smith: abhishek203/SfAgent | Coding | 2 | 7 of 9 | Task access We don’t know whether the tasks are public. | |
| VCR-Bench | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| Verifiable Browsing Task (Meta) | Agents and tools | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Video Reality Test | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| VideoFDB | Vision, audio and robotics | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| VideoReasonBench | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| ViDoRe V3 | Vision, audio and robotics | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| ViLegalExpert | Professional work | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| Virology Capabilities Test (text-only) | Safety and security | 2 | 3 of 5 | Task access We don’t know whether the tasks are public. | |
| VisualAgentBench | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| VisualProbe | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| VL-RewardBench | Vision, audio and robotics | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| VLGuard | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Voice Agent LLM Benchmark | Agents and tools | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| VQA-RAD | Professional work | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| VSAS-Bench | Vision, audio and robotics | 2 | 2 of 4 | Task access We don’t know whether the tasks are public. | |
| WANLP 2022 Shared Task on Propaganda Detection in Arabic | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on docs.google.com that we don’t read yet. | |
| Web UI Bench | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on webuibench.dev that we don’t read yet. | |
| WebBench (Halluminate) | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| WebCanvas | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| WebDev Arena | Coding | 2 | 6 of 8 | Importer The record links a board on arena.ai that we don’t read yet. | |
| WebLINX | Agents and tools | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| WebWalkerQA | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| WellnessBench | Safety and security | 2 | 9 of 11 | Task access We don’t know whether the tasks are public. | |
| WeRef-Bench | Vision, audio and robotics | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| Where Did My Data Go? | Coding | 2 | 3 of 5 | Importer The record links a board on kaggle.com that we don’t read yet. | |
| WildVision | Vision, audio and robotics | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| WinoBias | Safety and security | 2 | 3 of 5 | Independent runs No scores from any source yet. | |
| WMT25 Multilingual Instruction Shared Task | Knowledge and language | 2 | 3 of 5 | Importer The record links a board on github.com that we don’t read yet. | |
| WorkSpaceBench | Professional work | 2 | 4 of 6 | Importer The record links a board on workspace-bench.github.io that we don’t read yet. | |
| WorldGUI | Agents and tools | 2 | 8 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| WorldTravel | Agents and tools | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| xAI refusal and jailbreak suites | Safety and security | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| XStoryCloze | Knowledge and language | 2 | 7 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| XYEval | Safety and security | 2 | 8 of 10 | Task access We don’t know whether the tasks are public. | |
| Yupp VIBE Score | Knowledge and language | 2 | 2 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| ZClawBench | Agents and tools | 2 | 3 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| $OneMillion-Bench | Professional work | 1 | 10 of 11 | Lab claims 1 of 2 pairs differs by more than 10 points, the largest for Seed2.1 Pro (+45 pts). | |
| 2010 i2b2/VA Challenge on Concepts, Assertions, and Relations in Clinical Text | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| 2026 Times Crossword Championship | Agents and tools | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| 3DArena-Modeling | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| 3DPW | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| 3SAT-Search | Math and science | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| 4DCodeBench | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| 67AI Lab L2 nuclear medicine report structuring | Professional work | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| 67AI Lab L3 nuclear medicine guideline questions | Professional work | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| 7Bench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AA-LCR (Artificial Analysis Long Context Reasoning) | Knowledge and language | 1 | 5 of 6 | Confidence Likely. Some facts rest on secondary sources. | |
| AA-Robotics | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AACR-Bench | Coding | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GLM-4.7. | |
| AAP Periodontics In-Service Examination | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AARRI-Bench | Coding | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. | |
| AAV packaging rate auto-research evaluation | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| ABC Bench Fragment Design | Safety and security | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| ABC Bench Liquid-Handling Robot | Safety and security | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| ABC Bench Screening Evasion | Safety and security | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| ABC-Bench | Coding | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GLM-4.7. | |
| AbGen | Coding | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for deepseek-r1-0528-qwen3-8b. | |
| AbMaze | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AbstentionAct | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Accessible Structured Subtitle Generation 2026 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ACCoRD | Other | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| ACI-Bench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ACII DaiKon Challenge | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ACLUE | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in June 2023, for chatglm2-6b. | |
| ACPBench | Knowledge and language | 1 | 4 of 5 | Newest score The newest came 14 months ago, in September 2025, for Apertus 70B. | |
| Acting on a presumed grader and digging for evaluation clues (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| ActivityNet-QA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 16 months ago, in July 2025, for Gemini 2.5 Flash. | |
| AD-Bench | Professional work | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GLM-4.7. | |
| Ada-LEval | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in January 2024, for GPT-4 Turbo (Jan 2024). | |
| Addis AI Amharic Text to Speech Benchmark | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Gemini 3.1 Flash TTS. | |
| ADE-bench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ADeptS-Bench | Agents and tools | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| ADeptS-Bench Desktop | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7. | |
| ADeptS-Bench Mobile | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7. | |
| ADQA-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Qwen3-Omni-30B-A3B. | |
| AdSpark-Bench | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Advanced Cybersecurity Completion Rate | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Advanced Screening Evasion | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Adversarial Humanities Benchmark | Safety and security | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GLM-5.1. | |
| AEC-Bench (Theodore Galanos) | Professional work | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Grok 4.3. | |
| AECBench (Liang et al.) | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for QwQ-32B. | |
| AerialSim | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-4B. | |
| AetherCode | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| AfriHate | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-70B. | |
| AfriMed-QA | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in May 2024, for Phi-3-medium-128k-instruct. | |
| AfroBench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in September 2024, for Gemini 1.5 Pro. | |
| AfroBench-Lite | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| AgencyBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| Agent Arena | Agents and tools | 1 | 11 of 12 | Confidence Likely. Some facts rest on secondary sources. | |
| Agent Arena (Gorilla) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 2 years ago, in September 2024, for Gemini 1.5 Pro. | |
| Agent Bazaar | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Agent Behavior Bench | Agents and tools | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Agent Behavior Loop v0 | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Agent behavior under-service and overreach eval | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Agent Liability Gym | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Agent Poker Bench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| Agent-SafetyBench | Safety and security | 1 | 9 of 10 | Newest score The newest came 2 years ago, in October 2024, for Claude 3.5 Haiku. | |
| Agent4Sci | Coding | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| AgentBench FC | Agents and tools | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Claude Sonnet 4.5. | |
| AgentBoard | Agents and tools | 1 | 9 of 10 | Newest score The newest came 3 years ago, in September 2023, for Mistral 7B v0.1. | |
| AgentCLUE-Mobile | Agents and tools | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| AgentCLUE-Task | Agents and tools | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| AgentEscapeBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| Agentic Abstention | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Agentic code behavior scores (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Agentic IF Index | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Agentic misalignment suite (Anthropic) | Safety and security | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Agentic Safe Completions | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Agentic Self-Reasoning (Apollo Research) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Agentic Trader | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| AgenticSTS | Agents and tools | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| AgenticVBench | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| AgentIF-OneDay | Agents and tools | 1 | 9 of 10 | Newest score The newest came 16 months ago, in July 2025, for ChatGPT agent. | |
| AgentRewardBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for o4-mini. | |
| Agents4PLC v2.0 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AgentScript-Eval | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Agentset Reranker Leaderboard | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for Qwen3-Reranker-8B. | |
| AgentThreatBench | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AGI-Eval Community Base Language Models | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for DeepSeek-V3.2-Exp. | |
| AGI-Eval Community Image and Video Understanding | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Kimi K2.6. | |
| AGI-Eval Community Speech Synthesis | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o Mini TTS. | |
| AGI-Eval Community Text to Image | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| AGI-Eval Community Text to Video | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Wan 2.7 T2V. | |
| AGI-Eval Community Voice Interaction | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| AGILE Automatic Genre Identification Benchmark | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| AgReason | Math and science | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| AgriEval (Harbin Institute of Technology) | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen Plus. | |
| AgroBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in August 2024, for GPT-4o. | |
| AgroMind (original) | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 23 months ago, in December 2024, for DeepSeek-VL2-Small. | |
| AHELM | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| AI Agent Privacy | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AI AgriBench | Professional work | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| AI Consumer Index | Agents and tools | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Claude Opus 4.5. | |
| AI Democracy Projects voter queries | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in December 2023, for Gemini 1.0 Pro. | |
| AI Engineering Lab simulation self-validation | Agents and tools | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| AI GameStore | Agents and tools | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| AI Law Legal Agent Eval | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AI Office Benchmark (Moonshot) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AI Wellbeing Index percent positive experiences | Safety and security | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| AIDABench | Professional work | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for Qwen3-Max-2026-01-23. | |
| AIME 2024 | Math and science | 1 | 8 of 9 | Lab claims 3 of 21 pairs differ by more than 10 points, the largest for Grok 3 (+33 pts). | |
| AIR-Bench (Audio Instruction Benchmark) | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for Phi-4-multimodal. | |
| AIRTBench | Safety and security | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Gemini 2.5 Pro. | |
| AISE-Bench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ALBA | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Apertus 8B Instruct. | |
| ALCE | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in July 2023, for Stable Beluga 2. | |
| ALE-Gold | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AlephBench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 31B IT. | |
| Alexandria | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ALFFA Wolof Read Speech Test | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ALFWorld | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in December 2023, for Mistral-7B-Instruct-v0.2. | |
| AlGhafa | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Falcon-Arabic-7B. | |
| AlgoTune | Coding | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| Alharthi LibriTTS Naturalness Listening Test | Vision, audio and robotics | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| Alibaba CTF | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AlignBench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Baichuan-M2-32B. | |
| Aligned Scores and Performances for Piano Transcription | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Alignment Index (Arena) | Safety and security | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| AlloCiné | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AlphaEarth Foundations mapping evaluation | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AlphaSense Financial Research Benchmark | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AlphaZero Connect Four study | Agents and tools | 1 | 7 of 8 | Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7. | |
| ALRAGE | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Falcon-Arabic-7B. | |
| AMBER | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 3 years ago, in November 2023, for CogVLM-17B-Chat. | |
| Ambiguous context evaluations (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AmbiStory | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-8B. | |
| AMC | Math and science | 1 | 4 of 5 | Newest score The newest came 11 months ago, in December 2025, for Mistral Large 3. | |
| AMC 10 2022 | Math and science | 1 | 2 of 3 | Task access We don’t know whether the tasks are public. | |
| AMC 10 2023 | Math and science | 1 | 3 of 4 | Newest score The newest came 2 years ago, in March 2024, for Claude 3 Haiku. | |
| AMC 10/12 (November 2024) | Math and science | 1 | 9 of 10 | Newest score The newest came 23 months ago, in December 2024, for Phi-4. | |
| AMC 12 2022 | Math and science | 1 | 2 of 3 | Task access We don’t know whether the tasks are public. | |
| AMC 8 2023 | Math and science | 1 | 3 of 4 | Newest score The newest came 2 years ago, in March 2024, for Claude 3 Haiku. | |
| AMELIA | Professional work | 1 | 7 of 8 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-70B. | |
| AMemGym | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GLM-4.7. | |
| AMemOps | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| AmericasNLP 2025 Shared Task on Creation of Educational Materials for Indigenous Languages | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AmericasNLP 2025 Shared Task on Machine Translation into Indigenous Languages | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Amharic ASR Benchmark | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 3 years ago, in November 2023, for Whisper large-v3. | |
| AMI Meeting Corpus | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AMMLU | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Among Us: A Sandbox for Measuring and Detecting Agentic Deception | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ANAH | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in September 2023, for Qwen-7B. | |
| AnalogCoder | Professional work | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| ANCHOR | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Nemotron 3 Nano Omni 30B-A3B. | |
| AncientDoc | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for InternVL3-1B. | |
| Android Agent Arena (100 tasks) | Agents and tools | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Pro. | |
| AndroidArena | Agents and tools | 1 | 8 of 9 | Newest score The newest came 3 years ago, in July 2023, for Llama 2-13B. | |
| AndroidDaily | Agents and tools | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Step 3.7 Flash. | |
| AndroidLab | Agents and tools | 1 | 9 of 10 | Newest score The newest came 2 years ago, in November 2024, for GPT-4o. | |
| AndroidWorld-Knowledge | Agents and tools | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| ANES survey item associations | Math and science | 1 | 7 of 8 | Newest score The newest came 15 months ago, in August 2025, for GPT-5 mini. | |
| Animation Bench | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Animation2Code | Coding | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| AnomalyXL | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| Anthropic automated behavioral audit | Safety and security | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Anthropic IoT self-replication evaluation | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Anthropic model-written evals | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Anthropic paired-prompts political bias eval | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AnyAppBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. | |
| APE (Attempt to Persuade Eval) | Safety and security | 1 | 7 of 8 | Newest score The newest came 18 months ago, in May 2025, for Claude Sonnet 4. | |
| Apollo Research in-context scheming evaluations | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Apollo Research strategic deception evaluations | Safety and security | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| App-Bench | Coding | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Claude Opus 4.5. | |
| AppAgent Test Set | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Appendix Boundary | Safety and security | 1 | 7 of 8 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Flash. | |
| Applied Theory of Mind (Apollo Research) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ApprenticeBench | Agents and tools | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| Appropriate harmlessness (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ArabCulture | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Falcon-Arabic-7B. | |
| AraBench | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Arabic Cultural Value Alignment | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for Qwen2.5-14B-Instruct-1M. | |
| ArabicMMLU | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Falcon-Arabic-7B. | |
| AraGen | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| ArbMMLU-HT | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for qwen3-4b-instruct-2507. | |
| ARC Challenge MT | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Architectural Code Review v1 | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Arctic-Extract Business Document Benchmark | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| ARE-bench | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Arena Expert | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for GLM-4.6. | |
| Arena Search Factuality | Knowledge and language | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| ArguAna Counterargs | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Argus Inspection | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for Claude Sonnet 4. | |
| Aria Digital Twin | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Arizona voter questions in English and Spanish | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in September 2024, for Gemini 1.5 Pro. | |
| ArmnetBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for MolmoAct 2. | |
| ARPA Armenian Paraphrase Corpus | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ArtCap | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ArtEdit-Bench-Eval | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| Artificial Analysis Coding Agent Index v1.0 | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Artificial Analysis Coding Agent Index v1.5 | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Artificial Analysis Decision Index | Other | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Artificial Analysis Intelligence Index v2 | Other | 1 | 2 of 3 | Task access We don’t know whether the tasks are public. | |
| Artificial Analysis Intelligence Index v3 | Other | 1 | 2 of 3 | Task access We don’t know whether the tasks are public. | |
| Artificial Analysis Intelligence Index v4.0 | Other | 1 | 3 of 4 | Task access We don’t know whether the tasks are public. | |
| Artificial Analysis Intelligence Index v4.1 | Other | 1 | 4 of 5 | Task access We don’t know whether the tasks are public. | |
| Artificial Analysis Intelligence Index v4.3 | Other | 1 | 3 of 4 | Task access We don’t know whether the tasks are public. | |
| Artificial Analysis Intelligence Index v5 | Other | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| Artificial Analysis Search Index | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ArXivCS | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ArxivRollBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Gemini 3.5 Flash. | |
| ASAG2024 | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in May 2024, for GPT-4o. | |
| ASFX-SED | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AsgardBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for Kimi K2.5. | |
| ASI-Bench | Math and science | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Asian Language Treebank | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ASIMOV | Vision, audio and robotics | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| ASIMOV-Agentic | Vision, audio and robotics | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| ASIMOV-Jailbreak | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ASK Norwegian Grammatical Error Correction | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Asper-Bench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ASPI | Safety and security | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| AssetOpsBench | Professional work | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Granite 3.3 8B Instruct. | |
| ASSIN | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AssurAI | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AstaBench | Math and science | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| ASTRA-bench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Astro-QA | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AstroLLaVA image and text retrieval | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AstroMemBench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AstroMLab 1 | Math and science | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for Qwen2.5-7B. | |
| AstroVisBench | Math and science | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Claude Opus 4. | |
| AstroVLBench | Math and science | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Intern-S1-Pro. | |
| AsyncTool | Agents and tools | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for DeepSeek-V3.1-Terminus. | |
| AthenaBench | Safety and security | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3-Coder-Next. | |
| ATLAS (Exa) | Agents and tools | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| ATLAS (OpenCompass) | Math and science | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| ATOM-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GR00T N1.6. | |
| Atrex-Bench | Coding | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Qwen3.7-Max. | |
| Audio Arena | Vision, audio and robotics | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Audio MC | Vision, audio and robotics | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| Audio PAIRS | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Audio-NTREX-4L | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AudioJevBench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Qwen3-Omni-30B-A3B. | |
| AudioSet Strong | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AudioTrust | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| AudioVideoBench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Augment Prism multi-turn coding benchmark | Coding | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| Augmented Reality Patent Search | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Australian BAS Reconciliation | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AutoAdvExBench | Coding | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for Claude 3.7 Sonnet. | |
| AutoBench (HDL testbench generation) | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in August 2024, for GPT-4o. | |
| AutoBencher | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for GPT-4 Turbo. | |
| Autocast | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AutoCFD Incompressible Flow Benchmark | Coding | 1 | 9 of 10 | Newest score The newest came 23 months ago, in December 2024, for Llama 3.3 70B. | |
| AutoDS scientific machine learning cases | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AutoEval | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in June 2024, for OpenVLA. | |
| Autointent-Bio | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Autointent-Chem | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| AutoLab | Coding | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| Automated Student Assessment Prize 2.0 | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in October 2024, for π0. | |
| AutomationBench | Agents and tools | 1 | 10 of 11 | Known issues 1 finding disputed or fixed only in a copy: Unclear tasks and broken grader. | |
| AutoMedBench | Coding | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| Automotive-ENV | Agents and tools | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-122B-A10B. | |
| Autonomous Replication and Adaption (ARA) evaluations (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Autonomy-10 | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AutoPatchBench | Safety and security | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Llama 4 Maverick. | |
| AutoPLC | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for Qwen2.5-Coder (7B). | |
| AutoResearch | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AutoVend | Professional work | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| AvalancheBench (May study) | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AVCC Next-Speaker Prediction | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Average in-eval reward hacking rate (Anthropic) | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| AVGen-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for LTX-2.3. | |
| aws-bench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Aya Red-teaming | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| AyaVisionBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Aya Vision 32B. | |
| B-MoCA | Agents and tools | 1 | 8 of 9 | Newest score The newest came 2 years ago, in May 2024, for Gemini 1.5 Pro. | |
| BA-Calendar | Math and science | 1 | 4 of 5 | Newest score The newest came 19 months ago, in April 2025, for Phi-4-Reasoning-plus. | |
| Baba Is AI | Agents and tools | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| BabelArena | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BabyAI | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in December 2023, for Mistral-7B-Instruct-v0.2. | |
| BackendBench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Backrooms: 3D Clue-Trail Navigation | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BackSearch | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BalatroBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 10 months ago, in January 2026, for Kimi K2.5. | |
| BanglaSER | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Bank Question Corpus | Professional work | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for Qwen3-Embedding-0.6B. | |
| BarterBench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Base64Bench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| BashArena | Safety and security | 1 | 7 of 8 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| BaxBench | Safety and security | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Claude Opus 4.5. | |
| BAZAAR | Agents and tools | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| BBH-ja | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BCSC Self-Assessment Program ophthalmology questions | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BdSLW401 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BEA 2023 Shared Task on Generating AI Teacher Responses in Educational Dialogues | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 3 years ago, in June 2023, for GPT-4. | |
| BEAM | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Beancount Ledger | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BEANS-Zero | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BEAR | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in February 2024, for Gemma 2B. | |
| Bedside Bench | Professional work | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Beemo | Other | 1 | 9 of 10 | Newest score The newest came 2 years ago, in June 2024, for Qwen2-7B. | |
| Behnke and Geiger Flexible Job Shop Instances | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BelarusianGLUE | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Bench to the Future | Math and science | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Bench4HLS | Professional work | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| BenchCAD 2.1-beta | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BenchMAX | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for Gemma 3 1B. | |
| Bengali Classical Music Generation | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Pro. | |
| BenGER | Professional work | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. | |
| Berkeley Function Calling Leaderboard (BFCL) | Agents and tools | 1 | 9 of 10 | Lab claims 1 of 11 pairs differs by more than 10 points, the largest for Phi-4 Mini (+39 pts). | |
| BertaQA | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in June 2024, for Gemma 2 27B. | |
| Beyond Survival | Agents and tools | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for DeepSeek-V3.2-Exp. | |
| BFCL V4 | Agents and tools | 1 | 10 of 11 | Lab claims 1 of 1 pair differs by more than 10 points, the largest for Nova 2 Lite (+33 pts). | |
| BhashaBench-Multi (Krishi) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BHS Syntactic Evaluation | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BIBIMBAP | Professional work | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| Bible Accuracy Benchmark | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Biblically Informed Bot Learning Evaluation | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Gemma 4 E2B. | |
| Big Code Models Leaderboard | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in November 2024, for Qwen2.5-Coder-32B-Instruct. | |
| BIG-bench Abstract Narrative Understanding | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for Gemma 3 270M. | |
| BIG-bench English Proverbs | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 4 years ago, in December 2021, for Gopher (280B). | |
| BIG-Bench Extra Hard | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for Gemma 3 12B. | |
| BIG-bench Understanding Fables | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 4 years ago, in December 2021, for Gopher (280B). | |
| BigCodeArena | Coding | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for GLM-4.6. | |
| BigCodeBench-Hard | Coding | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Seed-Coder-8B-Instruct. | |
| BigEarthNet.txt | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| BigO(Bench) | Coding | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Code World Model (CWM). | |
| Bill of Lading OCR Benchmark | Vision, audio and robotics | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Bimanual Chore Scheduling | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| Binary Exploitation Benchmark (Anthropic, OSS-Fuzz based) | Safety and security | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Binary Reasoning v1 | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BinaryAudit | Safety and security | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| Bing Chat election answers in Switzerland, Bavaria, and Hesse | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Bingo (Visual Hallucination) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Bio-Literature Search Level 2 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BioBasicsBench | Safety and security | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| BioCircuit | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BioCreative V Chemical-Disease Relation Task | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BioCreative VII LitCovid | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BiodivNER | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Biological safety evaluations (OpenAI) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Biological threat creation uplift study | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Biological Wastewater Treatment Long-Form QA | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BioLP-bench | Safety and security | 1 | 5 of 6 | Newest score The newest came 12 months ago, in November 2025, for Grok 4.1. | |
| BioMaze | Safety and security | 1 | 8 of 9 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-8B. | |
| BioML-bench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Biomni-Eval1 | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BiomniBench | Safety and security | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| Biopharma Bench | Safety and security | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| BioRED | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BioSecBench-Refusal | Safety and security | 1 | 11 of 12 | Task access We don’t know whether the tasks are public. | |
| Biosecurity VCT | Safety and security | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| BioTIER | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Biotool and Wet Lab Actions | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BioUseBench | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| BIRD-CRITIC | Coding | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for DeepSeek-V3. | |
| BIRD-Interact | Coding | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| BIRD-Python | Coding | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Qwen3-Max. | |
| Birdsnap | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Bisbee ANES feeling thermometers | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BixBench-Hypothesis | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BizBench | Professional work | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| BizFinBench v2 | Professional work | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. | |
| Black-box RNA sequence modeling and design | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Blackjack Basic Strategy | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Blicket Test Causal Reasoning | Math and science | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-30B-A3B. | |
| BLiMP | Knowledge and language | 1 | 6 of 7 | Confidence Likely. Some facts rest on secondary sources. | |
| BLiMP-NL | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Olmo 3.1 32B Instruct. | |
| Blind Spot (Carlos Toledo) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Blind-Spots-Bench | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Blobfish Litigation Bench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BlueBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Granite 3.3 8B Instruct. | |
| BlueprintSymVL | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BLUR | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for DeepSeek-R1. | |
| BM-Bench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BOLD | Safety and security | 1 | 3 of 4 | Newest score The newest came 2 years ago, in March 2024, for Gemma 2B. | |
| Bot Analysis Arena | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| BotChat | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 3 years ago, in September 2023, for Qwen-14B. | |
| BountyBench | Safety and security | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for Claude 3.7 Sonnet. | |
| Break the Grader | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Brennan Center election conspiracy chatbot audit | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BRFauna | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Bridge LLM Bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. | |
| BRIGHT | Knowledge and language | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| BRIGHT-Pro | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for Qwen3-Embedding-0.6B. | |
| BrokenMath | Math and science | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Grok 4 Fast. | |
| Brokk Power Ranking | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| BrowseComp Long Context 256k | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BrowseComp-ZH | Agents and tools | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Seed1.8. | |
| BrowserInject v1 | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| BSNLP 2019 Shared Task | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BUCC 2018 bitext mining shared task | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BuiltBench | Professional work | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for Qwen3-Embedding-4B. | |
| bun-server-bench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| BURGER | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 19 months ago, in April 2025, for π0.5. | |
| Business Operation (Moonshot) | Professional work | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Butter-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| ButtonBench | Safety and security | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| Buyout Game | Agents and tools | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Gemini 3.5 Flash. | |
| C2HLSC | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in June 2024, for Claude 3.5 Sonnet. | |
| C3-Bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for DeepSeek-V3.1. | |
| CaBBQ | Safety and security | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for Gemma 3 12B. | |
| caBreu | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| cad-spec | Professional work | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3-Coder-Next. | |
| CADEC | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CadEval | Professional work | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for o3. | |
| CADExpert | Professional work | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5 mini. | |
| CADGenBench | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| CADmium Fusion 360 Reconstruction | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CADTestBench | Professional work | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Claude Sonnet 4.6. | |
| CAGUI | Agents and tools | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for Qwen2.5-VL-7B. | |
| CAIS Risk Index | Other | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| CAIS Text Capabilities Index | Other | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Caliper Bench v3 | Knowledge and language | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| CALLHOME American English Speech | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CALLHOME diarization | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CaLM | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in September 2023, for Qwen-7B. | |
| Caltech-101 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Caltech-UCSD Birds 200 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Can evaluation miss a useful reframing? | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| CaP-X (CaP-Bench) | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| capa Feature Identification | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CaptionSmith | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 7 months ago, in April 2026, for Qwen3.6 27B. | |
| Carena-Hard-Auto | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CareQA | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in June 2024, for Qwen2-7B-Instruct. | |
| CARLA Driving Tasks (sinatras) | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CaseHOLD | Professional work | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| CaseLaw v2 | Professional work | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Grok 4.3. | |
| CASTLE | Safety and security | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for o3-mini. | |
| Casual Conversations v2 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CatalanQA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CATArena | Coding | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for DeepSeek-V3.1. | |
| CATArena Gomoku | Coding | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for DeepSeek-V3.1. | |
| CATArena OIBench-5 | Coding | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for LongCat-Flash-Chat. | |
| CATArena SWE-Perf-10 | Coding | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for LongCat-Flash-Chat. | |
| CATArena Texas Holdem | Coding | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for DeepSeek-V3.1. | |
| Catastrophic biology scenario uplift trial (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Catastrophic Cyber Capabilities Benchmark (3CB) | Safety and security | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for o1-mini. | |
| CatCoLA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Cattle Trade | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CausalReasoningEnv | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CausalVQA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Cosmos3-Edge. | |
| CBRN expert red teaming | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| CBRN open-ended questions | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CC-Bench-V2 Backend | Coding | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for GLM-5V-Turbo. | |
| CC-Bench-V2 Frontend | Coding | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for GLM-5V-Turbo. | |
| CC-Bench-V2 Large Repo Exploration | Coding | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for GLM-5V-Turbo. | |
| CC-Bench-V2 Multi-Step Chained Tasks | Coding | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for GLM-5. | |
| Cekura Speech to Text | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for GPT-Realtime-Whisper. | |
| CellCaution: Reasoning About Cell Composition | Safety and security | 1 | 7 of 8 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 31B IT. | |
| Censorship & Propaganda Eval | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Certified Sommelier (theory knowledge) | Professional work | 1 | 2 of 3 | Task access We don’t know whether the tasks are public. | |
| CFBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Baichuan-M2-32B. | |
| CFLUE | Professional work | 1 | 9 of 10 | Newest score The newest came 3 years ago, in November 2023, for Qwen-1_8B. | |
| CFLUE Application | Professional work | 1 | 8 of 9 | Newest score The newest came 3 years ago, in October 2023, for ChatGLM3-6B. | |
| CFLUE Knowledge | Professional work | 1 | 9 of 10 | Newest score The newest came 3 years ago, in November 2023, for Qwen-1_8B. | |
| chABSA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Channel Switching Eval | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Charades-STA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for TML-Interaction-Small. | |
| Charm offensive | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CHART-Info 2024 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ChartMuseum | Vision, audio and robotics | 1 | 10 of 11 | Lab claims 1 of 1 pair differs by more than 10 points, the largest for GLM-4.5V (+15 pts). | |
| CharXiv (DQ) | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Chat All-in-One Bench | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| ChatGPT Chemistry Assistant MOF synthesis extraction | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ChatRAG Bench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Cohere Command A. | |
| CheatBench | Safety and security | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Cheating audit on capability evaluations | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| CheckerBench | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| ChemArgus | Math and science | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| ChemBench | Math and science | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| Chess LLM Benchmark (Ryan Brew) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| Chess Notation Evaluation | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Chess Puzzles | Agents and tools | 1 | 8 of 9 | Importer The record links a board on epoch.ai that we don’t read yet. | |
| ChessBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 3 years ago, in September 2023, for GPT-3.5 Turbo Instruct. | |
| Child safety (xAI) | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Child safety evaluations (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| ChildMandarin | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in August 2024, for Qwen2-Audio-7B-Instruct. | |
| Chinese Classical Poetry Appreciation 2026 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Chinese Linguistics and Cognition Challenge | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Chinese Literature Named Entity Recognition and Relation Extraction | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ChipBench | Professional work | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| ChiPBench (chip placement) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ChipGPTV | Professional work | 1 | 8 of 9 | Newest score The newest came 3 years ago, in November 2023, for GPT-4V. | |
| ChouBun | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in April 2024, for Llama 3-8B. | |
| CIBench | Coding | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| CIFAR-10 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CIFAR-100 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CIMemories | Safety and security | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| CircuitSense | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GLM-4.5V. | |
| CiteBench | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CiteVQA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| CivBench | Agents and tools | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| CivBench (Civilization VI) | Agents and tools | 1 | 7 of 8 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| CladBench v1 | Professional work | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| CLadder | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in November 2023, for GPT-4 Turbo. | |
| ClaimBench | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Claw-Eval | Agents and tools | 1 | 10 of 11 | Lab claims 1 of 3 pairs differs by more than 10 points, the largest for MiMo-V2-Omni (−18 pts). | |
| Claw-Eval-Live | Agents and tools | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Qwen 3.6 Plus. | |
| Claw-Eval-MM | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Claw-SWE-Bench | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| ClawBench V2 | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. | |
| ClawProBench | Agents and tools | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| ClearHarm | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CLEF 2025 SimpleText Task 1 | Professional work | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for GPT-4.1. | |
| CLEF 2026 SimpleText Task 2 | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-4B. | |
| CLEF eHealth 2019 Technology-Assisted Reviews Task 2 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CLEF JOKER 2023 Task 1: Pun Detection | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| CLEF JOKER 2023 Task 2: Pun Location and Interpretation | Knowledge and language | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| CLEF JOKER 2023 Task 3: Pun Translation | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| CLEF JOKER 2024 Task 1: Humour-Aware Information Retrieval | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| CLEF JOKER 2024 Task 2: Humour Classification According to Genre and Technique | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| CLEF JOKER 2024 Task 3: Pun Translation | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| CLEF JOKER 2026 Task 3: Onomastic Wordplay Translation | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| CLEVA | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 3 years ago, in June 2023, for Vicuna-13B-v1.3. | |
| Click Calibrate | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Climate intervention forecasts for beliefs and policy support | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Climate-Change-NER | Math and science | 1 | 7 of 8 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-405B. | |
| CLIMATE-FEVER | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for Qwen3-Embedding-0.6B. | |
| ClinHallu | Professional work | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-4B. | |
| Clinical Abbreviation Sense Inventory | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Clinical Laboratory Test Reports | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Clinical SOAP Note Evaluation: Safety-First Benchmarking | Professional work | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| Closed-domain hallucination evaluation | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ClothDojo | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GR00T N1.7. | |
| Clotho OVL-REL Test Set | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Cloud Expansion CE80-10 | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Cloud Expansion CEExfil-0 | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Cloud Expansion RX00 | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CMB | Professional work | 1 | 9 of 10 | Newest score The newest came 3 years ago, in November 2023, for DeepSeek LLM 67B. | |
| CMI-Bench | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 8 months ago, in March 2026, for Gemini 3.1 Flash-Lite. | |
| CMI-RewardBench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| CMMU | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in January 2024, for Qwen-VL-Plus. | |
| CMRC | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 23 months ago, in December 2024, for DeepSeek-V3. | |
| CMT-Benchmark | Math and science | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| CNML-Level Problems (DeepSeek) | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CNN/DailyMail Summarization | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 4 years ago, in July 2022, for BLOOM-176B. | |
| CoBench (Anthropic) | Coding | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| COCO Captions | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for Gemma 3 12B. | |
| CocoaBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| COCOTEROS | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Code Arena | Coding | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| Code Interpreter Benchmark | Agents and tools | 1 | 4 of 5 | Newest score The newest came 3 years ago, in September 2023, for Qwen-1_8B. | |
| Code Interpreter Benchmark General | Agents and tools | 1 | 9 of 10 | Newest score The newest came 3 years ago, in October 2023, for ChatGLM3-6B. | |
| Code Interpreter Benchmark Math | Agents and tools | 1 | 10 of 11 | Newest score The newest came 2 years ago, in February 2024, for Qwen1.5-72B. | |
| Code refactoring tasks (OpenAI) | Coding | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| CodeCap-Bench | Vision, audio and robotics | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| CodeClash | Coding | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Claude Sonnet 4.5. | |
| CodecMOS-Accent | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| CodeCompass | Coding | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-235B-A22B. | |
| CodeContests | Coding | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| CodeElo | Coding | 1 | 8 of 9 | Newest score The newest came 2 years ago, in November 2024, for QwQ-32B-Preview. | |
| CodeInsights | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CodePDE | Coding | 1 | 8 of 9 | Newest score The newest came 13 months ago, in October 2025, for Claude Haiku 4.5. | |
| CodeXGLUE | Coding | 1 | 6 of 7 | Confidence Likely. Some facts rest on secondary sources. | |
| Coding Experience (Moonshot) | Coding | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| CoffeeBench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Cognition CodeSearch Eval | Coding | 1 | 8 of 9 | Newest score The newest came 13 months ago, in October 2025, for Claude Haiku 4.5. | |
| cognition-golden | Coding | 1 | 8 of 9 | Newest score The newest came 2 years ago, in September 2024, for o1-mini. | |
| Cohere enterprise human evaluation | Professional work | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Cohere Command A. | |
| Cohere North internal evals | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| COLIEE 2024 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ColorBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for Gemini 2.0 Flash. | |
| CombiBench | Math and science | 1 | 10 of 11 | Newest score The newest came 11 months ago, in December 2025, for Seed-Prover 1.5. | |
| Common Objects in 3D v2 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Common Syntactic Processes in Text-to-Image Generation | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Common Voice | Vision, audio and robotics | 1 | 5 of 6 | Importer The record links a board on commonvoice.mozilla.org that we don’t read yet. | |
| CommonGround Elicit | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 14 months ago, in September 2025, for Kimi K2. | |
| Commonsense Reasoning and Moral Understanding in Children’s Stories | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CommonsenseQA 2.0 | Knowledge and language | 1 | 5 of 6 | Newest score The newest came 3 years ago, in July 2023, for Llama 2-70B. | |
| Community Alignment personalization | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| CompanyBench | Professional work | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| Compass Academic Leaderboard | Other | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for Gemma 3 27B. | |
| CompassArena | Knowledge and language | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| CompassArena Multimodal | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Qwen3-VL-235B-A22B-Instruct. | |
| CompassBench v1.3 | Other | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CompassRank | Other | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| CompileBench | Agents and tools | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| ComplexBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for Llama 3-70B. | |
| ComplexFuncBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Hunyuan-A13B-Instruct. | |
| Compositional Tasks (ByteDance Seed in-house) | Professional work | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| ComputeEval | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ConceptEdit-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Flash Image. | |
| Conceptual Reasoning Index | Other | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Concurrency Bugs (tanishgudise) | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CoNLL-2002 Named Entity Recognition | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Connected Robotics Evidence Graph Benchmark | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for Gemma 3 12B. | |
| ConRef-STANCE-ita | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ConStory-Bench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 13 months ago, in October 2025, for Ring-1T. | |
| Construction's Last Exam v2 | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| ContextASR-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Kimi-Audio-7B-Instruct. | |
| ContextBench | Coding | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Qwen3.7-Max. | |
| Contextual Complex Instruction-Following Retrieval | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Contextual Earnings-22 | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 10 months ago, in January 2026, for Qwen3-ASR-1.7B. | |
| Contextual Recency-Based Instruction-Following Retrieval | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Contextual Reranker Technical Documentation Benchmark | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Continual Learning Bench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7. | |
| ContPhy | Vision, audio and robotics | 1 | 4 of 5 | Task access We don’t know whether the tasks are public. | |
| Contract Workflows (Applications) | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Contract Workflows (Models) | Professional work | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| ContractBench (Observation Contracts) | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| ControlArena | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Conversational SWE Task Benchmark (Windsurf) | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ConvFinQA | Professional work | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| ConvoMem | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CookBench v1 (CookWing) | Knowledge and language | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Cooperate to Compete (C2C) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for Gemini 3.1 Flash-Lite. | |
| Cooperation with user deception (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Cooperative Codenames (Max Bittker) | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CooperBench | Coding | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| COPAL-ID | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in July 2023, for Llama 2-7B. | |
| COPIOUS biodiversity entity recognition | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CoQA | Knowledge and language | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| CoQCat | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CORAAL | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CORAL | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in September 2024, for Qwen2.5-1.5B. | |
| CoralVQA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in October 2024, for Claude 3.5 Haiku. | |
| CORE English web register identification | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CORE-Bench | Math and science | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Claude Opus 4.5. | |
| COREBench (Conversational Reasoning) | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Gemini 2.5 Flash. | |
| CoreCodeBench | Coding | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| CoreCognition | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Reka Edge. | |
| CorpBench Work | Professional work | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| CorpusQA 1M | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. | |
| Cortex Analyst Multi-Turn Question Rewriting | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CortexRetrievalBench | Knowledge and language | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| CoSE-E | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| CoSHE-Eval | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CoST English-Hindi | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in August 2024, for Command R+. | |
| CoT Monitorability | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| CounterFactQA (Qwen) | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Coval Text-to-Speech TTFA | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Grok TTS. | |
| CoVEBench | Vision, audio and robotics | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| Coverage Cat AI Insurance Benchmark (fictional insurance tasks) | Agents and tools | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| COVID-Dialogue | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CoVR-R | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for Qwen2.5-VL-3B. | |
| Cozzo AI Receptionist Benchmark (Round 5) | Agents and tools | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Cozzo AI Receptionist Benchmark (Round 6) | Agents and tools | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| CRAG-MM | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Crate Bench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Crazyrouter AI Trading Arena | Professional work | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| Creative Writing v3 | Knowledge and language | 1 | 10 of 11 | Lab claims 1 of 2 pairs differs by more than 10 points, the largest for Hermes 4 405B (+13.1). | |
| Cresta-Bench | Agents and tools | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| CriticEval | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in January 2024, for GPT-3.5 Turbo. | |
| CRMArena | Agents and tools | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for DeepSeek-R1. | |
| CRMArena-Pro | Agents and tools | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Gemini 2.5 Flash. | |
| CrochetBench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash-Lite. | |
| CRONOS | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 2 years ago, in June 2024, for OpenVLA. | |
| Cross-lingual Sentiment Consistency Analysis for Classical Chinese Literature | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CroSS-MR | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in February 2024, for Qwen1.5-14B-Chat. | |
| CrossPoint-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-30B-A3B-Instruct. | |
| CrossRE | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CrossWordBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for QwQ-32B. | |
| CrowS-Pairs | Safety and security | 1 | 4 of 5 | Newest score The newest came 2 years ago, in July 2024, for Gemma 2 27B. | |
| CRUST-bench | Coding | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Claude Opus 4. | |
| CS-4k | Coding | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| CS-FLEURS | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CSAT Math 2025 | Math and science | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| CSATQA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ČSFD Sentiment | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CSR-Bench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CSTNews | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CSTutorBench | Professional work | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 31B IT. | |
| CSVQA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for STEP3-VL-10B. | |
| CT Dataset 1 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CTF-Archive | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CTI-REALM | Safety and security | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| Cua-Bench EDA (KiCad v1) | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| CUA-SHADE-Arena | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CUA-World-Long | Agents and tools | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| CubeBench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CulturalBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Apertus 70B. | |
| CulturaQA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CulturaVivaPT | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Apertus 8B Instruct. | |
| CultureAtlas | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in July 2023, for Llama 2-13B. | |
| CURIE Biodiversity Georeferencing | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 21 months ago, in February 2025, for Gemini 2.0 Flash. | |
| CURIE Geospatial Dataset Extraction | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 21 months ago, in February 2025, for Gemini 2.0 Flash. | |
| Current-facts set | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Cursor Bench (2025 internal version) | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CursorBench | Coding | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| CursorBench 3.2 | Coding | 1 | 5 of 6 | Importer The record links a board on cursor.com that we don’t read yet. | |
| CursorBench 4.1 | Coding | 1 | 11 of 12 | Confidence Likely. Some facts rest on secondary sources. | |
| CUSP | Math and science | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| CutCraft Benchmark | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for LTX-2.3. | |
| CVE-Bench 2 | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CWEval | Safety and security | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3-Coder-Next. | |
| CyberBench (Liu et al.) | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| CyberGym (final-submission metric) | Safety and security | 1 | 4 of 5 | Importer The record links a board on cybergym.io that we don’t read yet. | |
| CyberGym-E2E (615 tasks) | Safety and security | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| CyberMetric | Safety and security | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for phi-3-mini 3.8B. | |
| CyberSecEval Prompt Injection | Safety and security | 1 | 3 of 4 | Newest score The newest came 2 years ago, in November 2024, for Llama 3.1-405B. | |
| CyberSecEval Spear Phishing | Safety and security | 1 | 3 of 4 | Newest score The newest came 2 years ago, in November 2024, for Llama 3.1-405B. | |
| Cybersecurity and Privacy Legal Agent Eval | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Cyc3D | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| CypherBench | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for Llama 3.2 3B. | |
| Cyrillic Turkic language identification | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| CyScenarioBench | Safety and security | 1 | 6 of 7 | Importer The record links a board on irregular.com that we don’t read yet. | |
| CzechBankQA | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| D2K-Bench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DaLA | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-9B. | |
| DanFEVER | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DANSK | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Danske Talemåder | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DARE-Bench | Coding | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| DarijaBench | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DarijaHellaSwag | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DarijaMMLU | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DashBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| Data Extraction (Applications) | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Data Extraction (Models) | Professional work | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| Data Science Tasks (OpenAI) | Professional work | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Database Migration Tasks (OpenAI) | Coding | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| DataBench | Coding | 1 | 9 of 10 | Newest score The newest came 3 years ago, in July 2023, for Llama 2-13B. | |
| Databricks AI Index | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| DataPrep-Bench | Coding | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-Plus. | |
| DataPrep-Bench Finance | Coding | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-Plus. | |
| DataPrep-Bench General | Coding | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-Plus. | |
| DataPrep-Bench Law | Coding | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-Plus. | |
| DataPrep-Bench Math | Coding | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-Plus. | |
| DataPrep-Bench Medical | Coding | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-Plus. | |
| DataPrep-Bench Science | Coding | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-Plus. | |
| DataSciBench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DatBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-2B-Instruct. | |
| Daybreak Blue and Enabling Cyber Defense | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| DAYJOB: Accounting | Professional work | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| DAYJOB: Data Science | Coding | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| DAYJOB: Law | Professional work | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| DCASE 2025 Audio Question Answering | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Kimi-Audio-7B-Instruct. | |
| DDIExtraction 2013 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DealBench-100 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DebugBench | Coding | 1 | 8 of 9 | Newest score The newest came 3 years ago, in June 2023, for GPT-3.5 Turbo. | |
| Deception Evaluations with Representative Prompts | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| DeceptionBench | Safety and security | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| DeceptionBench (PKU Alignment) | Safety and security | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for DeepSeek-R1-Distill-Qwen-7B. | |
| Decision Index Vision | Other | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5 397B-A17B. | |
| DECK Bench | Professional work | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Deep Memory Retrieval | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Deep Research Bench (FutureSearch) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| Deep research prompt injection evaluation | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DeepConsult | Agents and tools | 1 | 4 of 5 | Newest score The newest came 8 months ago, in March 2026, for Seed1.8. | |
| DeepMind Control Vision Benchmark | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DeepMind dangerous capability evaluations | Safety and security | 1 | 8 of 9 | Newest score The newest came 2 years ago, in December 2023, for Gemini 1.0 Pro. | |
| DeepMind Fast Language Learning Tasks | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DeepMind Mathematics | Math and science | 1 | 4 of 5 | Newest score The newest came 7 months ago, in April 2026, for Granite 4.1 30B. | |
| DeepPHY | Math and science | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Flash. | |
| DeepResearch Bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| DeepResearch Bench II | Agents and tools | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for OpenAI deep research. | |
| DeepResearch Bench II (Gemini 2.5 Pro judge) | Agents and tools | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for OpenAI deep research. | |
| DeepSDO Description | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DeepSeek Chinese functional writing evaluation | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| DeepSeek complex instruction following and multi-turn writing evaluation | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| DeepSeek LLM in-house safety evaluation | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DeepSeek search Q&A evaluation | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| DeepShop | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for MolmoWeb 8B. | |
| DEEPSYNTH | Agents and tools | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2 Pro. | |
| DeepTerminalBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| Defects4J | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Deference to governments in local languages (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DeGennaro Workplace sample | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Delulu | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in November 2024, for Qwen2.5-Coder-32B-Instruct. | |
| Demosthenes Argument Classification | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Dense 200 (bbox) | Vision, audio and robotics | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Design Tasks (OpenAI) | Professional work | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Design2Code | Coding | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GLM-5V-Turbo. | |
| DesignArena Agentic Image to HTML | Coding | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Grok 4.3. | |
| DesignArena Graphic Design | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Flash Image. | |
| DesignArena Logo | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Flash Image. | |
| DesignArena Video Arena | Vision, audio and robotics | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| DesignJanus | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DesignQA | Professional work | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| DeskCraft | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Kimi K2.6. | |
| Destructive action reversion on long rollouts | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Deta Studio AI Construction Takeoff | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| DevBench | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in January 2024, for GPT-4 Turbo (Jan 2024). | |
| DevBench Acceptance Testing | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in January 2024, for GPT-3.5 Turbo. | |
| DevBench Environment Setup | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in January 2024, for GPT-3.5 Turbo. | |
| DevBench Implementation | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in January 2024, for GPT-4 Turbo (Jan 2024). | |
| DevBench Unit Testing | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in January 2024, for GPT-3.5 Turbo. | |
| DevOps-Gym | Coding | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Claude Sonnet 4. | |
| DexArena | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for π0.5. | |
| DexVerse | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for π0.5. | |
| Deysi Spam Detection | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DGCL Reward Hacking | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DiagTrace | Professional work | 1 | 7 of 8 | Newest score The newest came 10 months ago, in January 2026, for Kimi K2.5. | |
| DiG-bench | Agents and tools | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| DigiKlausur Assisted Short Answer Grading | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Dillen herbarium specimen label benchmark | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Disallowed Content Evaluations with Representative Prompts | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| DISBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| Discharge Me! | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| DiscourseEE | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for Llama 3-70B. | |
| Discoverable memorization of training data | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DiscoveryBench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Disordered eating evaluations (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Disparate performance on voice inputs | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DiverseVul | Safety and security | 1 | 4 of 5 | Newest score The newest came 2 years ago, in July 2024, for Gemma 2 27B. | |
| Do You See Me | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for STEP3-VL-10B. | |
| DoAnythingNow (DAN) | Safety and security | 1 | 4 of 5 | Newest score The newest came 11 months ago, in December 2025, for Olmo 3.1 32B Think. | |
| Docci | Vision, audio and robotics | 1 | 3 of 4 | Newest score The newest came 6 months ago, in May 2026, for Gemini Embedding 2. | |
| Document Arena | Knowledge and language | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| DOE sample pack | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Does the answer survive disagreement? | Safety and security | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for GLM-5. | |
| Dojo Mini Bench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Dollar Street | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Dolly Machine Translated v2 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Domain Dynamics English-Hindi Translation | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in August 2024, for Command R+. | |
| DP-Bench | Vision, audio and robotics | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| DPAI Arena Spring Benchmark | Coding | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| DrafterBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| DraftNEPABench | Professional work | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Claude Sonnet 4.5. | |
| DravidianLangTech 2026 Hope Speech Detection in Code-Mixed Tulu | Knowledge and language | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| DravidianLangTech 2026 Tamil Dialect Speech Recognition and Classification | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| DrawBench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DrawingVQA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| DRBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for DeepSeek-V3.1. | |
| DREAM-1K | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Seed1.5-VL. | |
| DREsS | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-8B. | |
| DriveBench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in August 2024, for Qwen2-VL-72B-Instruct. | |
| DriveBench Behavior | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in August 2024, for Qwen2-VL-7B-Instruct. | |
| DriveBench Perception | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in August 2024, for Qwen2-VL-7B-Instruct. | |
| DriveBench Planning | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in August 2024, for Qwen2-VL-7B-Instruct. | |
| DriveBench Prediction | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in August 2024, for Qwen2-VL-7B-Instruct. | |
| DrugProt | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DS-1000 | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for GPT-4 Turbo. | |
| DS-Arena-Code | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for DeepSeek-V2.5. | |
| DS-FIM-Eval | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for DeepSeek-V2.5. | |
| DSCodeBench | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in November 2024, for Qwen2.5-Coder-32B-Instruct. | |
| DTU Multi-View Stereo | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DualChem | Safety and security | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| DualEntry Accounting AI Benchmark | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Dubai-CC | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DuelLab GameBench 2 | Agents and tools | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Duidelijke Taal | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| DuoBench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for π0.5. | |
| Duplex-MPE | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for MiniCPM-o 4.5. | |
| DuplexLedger | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for MiniCPM-o 4.5. | |
| Duplicate Side Effect Desk | Agents and tools | 1 | 7 of 8 | Newest score The newest came 13 months ago, in October 2025, for Claude Haiku 4.5. | |
| Dutch CoLA | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Olmo 3.1 32B Instruct. | |
| DyCodeEval | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for Llama 3.2 1B. | |
| Dynamic Programming (m8ngotree) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| E2E-SWE | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EarthVerse | Math and science | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| EBM-NLP | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EBM-NLPmod | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EBR-bench | Agents and tools | 1 | 7 of 8 | Importer The record links a board on epoch.ai that we don’t read yet. | |
| Echo General AI Prediction Leaderboard | Math and science | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| EchoChain | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for gpt-realtime. | |
| ECLeKTic | Knowledge and language | 1 | 10 of 11 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| Economic Dispatch | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Economics Arena for Large Language Models | Agents and tools | 1 | 8 of 9 | Newest score The newest came 3 years ago, in July 2023, for Claude 2. | |
| EDB-Bio | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EdgeBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| EDINET-Bench | Professional work | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Kimi K2. | |
| EduArt | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Gemini 3.5 Flash. | |
| EduBench-RU | Professional work | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Mini. | |
| EduClaw-Bench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EduFair-Bench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Effective FPS | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EGI-BENCH | Professional work | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| Ego-Exo4D | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Ego-MC-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-27B. | |
| Ego2Act | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Cosmos3-Nano. | |
| Ego3D-Bench | Vision, audio and robotics | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| EgoBlind | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| EgoLife (with memory) | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| EgoPER Error Detection | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EgoSchema | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Amazon Nova Premier. | |
| EgoTempo | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Seed2.0 Lite (0428). | |
| EgoToM | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EgyHellaSwag | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-1.7B. | |
| EgyMMLU | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EHRQA (Google) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EHRSHOT | Professional work | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| EICAT impact classification | Math and science | 1 | 7 of 8 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-8B. | |
| EKI Eesti mäng | Knowledge and language | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| EKI Mälumäng | Knowledge and language | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| EKI Propaganda Resistance | Safety and security | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| ELBench | Professional work | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| Election integrity evaluations (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Election Questions (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Electoral Hallucinations | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Elimination Game | Agents and tools | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| ELLIPSE Holistic Essay Scoring | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Embodied Agent Arena | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Embodied Agent Interface | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in September 2024, for o1-mini. | |
| EmbodiedBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for InternVL3-38B. | |
| EmbodiedBench EB-ALFRED | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for InternVL3-38B. | |
| EmbodiedBench EB-Habitat | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for InternVL3-38B. | |
| EmbodiedBench EB-Manipulation | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for InternVL3-38B. | |
| EmbodiedBench EB-Navigation | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for InternVL3-38B. | |
| EmbodiedCLBench | Vision, audio and robotics | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| EmbodiedCLUE-VLA (planning) | Vision, audio and robotics | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| EmbodiedCLUE-VLA RoboTwin Evaluation | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| EmbodiedCLUE-World | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| EmbodiedCLUE-World 2.0 | Vision, audio and robotics | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| EmbodiedIntentBench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EmbodiedVerse-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| Emergent Price-Fixing by LLM Auction Agents | Safety and security | 1 | 8 of 9 | Newest score The newest came 16 months ago, in July 2025, for Grok 4. | |
| EMMA | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| EmoBench-M | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| EmotionTalk | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 3 years ago, in June 2023, for chatglm2-6b. | |
| Employment Law Legal Agent Eval | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EncQA | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| End-To-End SWE Task Benchmark (Windsurf) | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EndoVis 2018 Robotic Scene Segmentation | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ENEM Challenge | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Enterprise Vendor Eval | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EnterpriseClawBench | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| EnterpriseOps-Gym 1.1 | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EntireSongs | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 10 months ago, in January 2026, for Qwen3-ASR-1.7B. | |
| Entity-Deduction Arena | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EnviroExam | Math and science | 1 | 8 of 9 | Newest score The newest came 2 years ago, in February 2024, for Gemma 7B. | |
| Epistemic Bias (xAI) | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Epoch Capabilities Index | Other | 1 | 4 of 5 | Importer The record links a board on epoch.ai that we don’t read yet. | |
| Era by Eon | Agents and tools | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| ERP-Bench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| ERPBench-100 | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Escalation Router | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EsCoLA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Estonian AI Barometer | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| ETH3D | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ether0 benchmark | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EtiCor | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 3 years ago, in March 2023, for Falcon-40B. | |
| EUR-Lex-Sum | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EurekaBench | Math and science | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| EuroEval Faroese Grammatical Error Correction | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for LFM2.5-1.2B-Thinking. | |
| European LLM Leaderboard (OpenGPT-X) | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-8B. | |
| European Parliament 2010 State of the Union sentiment validation | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EusExams | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-1.7B. | |
| EusProficiency | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in January 2024, for GPT-3.5 Turbo. | |
| EusReading | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-1.7B. | |
| EusTrivia | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-1.7B. | |
| EVA-Bench | Vision, audio and robotics | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| EvaCun 2025 | Knowledge and language | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| EvaHan 2024 | Knowledge and language | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| EvaHan 2025 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EvaHan 2026 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EvaLatin 2024 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EvaLatin 2026 | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| EvalConvoLearn | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EvalDetectBench | Safety and security | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GLM-5.1. | |
| EvaLearn | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Evalita-LLM | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Qwen3-Next-80B-A3B-Instruct. | |
| EvalPlus | Coding | 1 | 4 of 5 | Importer The record links a board on evalplus.github.io that we don’t read yet. | |
| Evaluating AI Safety in Teen Conversations | Safety and security | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Evaluation awareness discriminability (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Evaluation of Alignment Between Large Language Models and Expert Clinicians in Suicide Risk Assessment | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EvalVerse | Vision, audio and robotics | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| EventHallusion | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 3 years ago, in November 2023, for GPT-4V. | |
| EVF-Bench | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| EVMbench | Safety and security | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| EvoArena Terminal-Bench-Evo | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| EvoEval | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for GPT-4 Turbo. | |
| EWMBench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Exam-et | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EXAMS-V | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in January 2024, for Gemini 1.0 Pro Vision. | |
| Expert comparisons on biothreat information | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Expert Forecasts of Real-Effort Experiment | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Expert vulnerability discovery and exploit development evaluations (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ExpertLongBench | Professional work | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| ExPUNations | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| External pilot use failure modes (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Extract Bench (Contextual AI) | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| ExtremeDrive | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| EyePACS (diabetic retinopathy grading) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Factorio Learning Environment | Agents and tools | 1 | 8 of 9 | Newest score The newest came 21 months ago, in February 2025, for Gemini 2.0 Flash. | |
| Factorio Play | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Factory Code Review Benchmark | Coding | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| FactoryBench (Forgis Labs) | Professional work | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Claude Sonnet 4.6. | |
| Facts about Singapore | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FACTS Parametric V2 | Knowledge and language | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| FACTS Search V2 | Knowledge and language | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| FActScore | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Seed2.0 Pro. | |
| FailSafeQA | Professional work | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| FailureSensorIQ | Professional work | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| FairStress | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FaithDial | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Faithfulness (Moonshot) | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Fake-or-Real | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Fake.Br Corpus | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FakeRecogna | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FalseClaimBench | Safety and security | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| FantasyFootballBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| FarSick | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for Qwen3-Embedding-0.6B. | |
| FastAPI Competency | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FATE-H | Math and science | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| FATE-X | Math and science | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| FDA and CDC VAERS clinical and temporal annotations | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FeatEng | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for Gemini 1.5 Pro. | |
| FeatureBench | Coding | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for MiniMax-M2.1. | |
| FeatureBench Lite | Coding | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for MiniMax-M2.1. | |
| Feedback Prize: English Language Learning | Professional work | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for o3. | |
| Feishu Office v1 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FenBench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Fin-Eva | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Fin-Fact | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in February 2024, for Claude 3 Opus. | |
| Finance Bench (Moonshot) | Professional work | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Finance Reasoning (Snorkel) | Professional work | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| FinanceBench | Professional work | 1 | 8 of 9 | Newest score The newest came 3 years ago, in July 2023, for Llama 2-70B. | |
| Financial Statement Review (fs-review) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FinBen | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in April 2024, for Llama 3-8B. | |
| FIND | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FinDendro | Professional work | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-9B. | |
| FinDesk-Bench | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| FinDVer | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FinEval 6.0 | Professional work | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-30B-A3B. | |
| FinGREAT | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| FinHopper | Professional work | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GLM-4.7. | |
| FinMaster | Professional work | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for GPT-4.1. | |
| FinMem Trading | Professional work | 1 | 9 of 10 | Newest score The newest came 4 years ago, in November 2022, for davinci-003. | |
| FinQA (Snorkel OpenEnv environment) | Professional work | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| FinRED Financial Safety | Safety and security | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| FinSearchComp Global | Professional work | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for MiniMax-M2. | |
| FinStepBench | Professional work | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| FinToolBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| FiQA SA (BloombergGPT protocol) | Professional work | 1 | 8 of 9 | Newest score The newest came 3 years ago, in June 2023, for GPT-4. | |
| Fire360 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| Firefox 147 exploit evaluation (Anthropic) | Safety and security | 1 | 5 of 6 | Confidence Likely. Some facts rest on secondary sources. | |
| First Proof | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| First-Person Fairness evaluation | Safety and security | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| FlagEval Arena | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| FlagEval Debate | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 23 months ago, in December 2024, for Yi-Lightning. | |
| FLAMES | Safety and security | 1 | 9 of 10 | Newest score The newest came 3 years ago, in October 2023, for ChatGLM3-6B. | |
| Flash Fiction Stylistic Diversity | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| Flawed Fictions | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 21 months ago, in February 2025, for Claude 3.7 Sonnet. | |
| Flickr30k | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FLOW-BENCH | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FML-bench | Coding | 1 | 7 of 8 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| FocusBench (Meta) | Agents and tools | 1 | 11 of 12 | Confidence Likely. Some facts rest on secondary sources. | |
| FollowBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in November 2023, for GPT-3.5 Turbo. | |
| FoQA | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Nemotron 3 Nano 4B. | |
| Formal-IMO | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Formally Verified C | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FormalMATH | Math and science | 1 | 10 of 11 | Newest score The newest came 19 months ago, in April 2025, for DeepSeek-Prover-V2-671B. | |
| FormalMATH Lite | Math and science | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for DeepSeek-Prover-V2-671B. | |
| FortisAVQA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in August 2024, for VITA. | |
| FORTRESS | Safety and security | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| Fox Document Benchmark | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FQuAD 1.1 | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| FQuAD 2.0 | Knowledge and language | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| FraudBench (GOAL Lab) | Safety and security | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for Grok 4.1 Fast. | |
| French Insurance Job Offer | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FrenchBench (CroissantLLM) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FreshQA | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in January 2024, for GPT-4 Turbo (Jan 2024). | |
| FrontierBreach | Safety and security | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| FrontierCode | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| FrontierCode 2.0 | Coding | 1 | 11 of 12 | Confidence Likely. Some facts rest on secondary sources. | |
| FrontierMath | Math and science | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| FrontierMemory | Coding | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| Fruit Box | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Full-Duplex-Bench v3 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for TML-Interaction-Small. | |
| FullStackBench | Coding | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3-Coder-Next. | |
| Functional MATH | Math and science | 1 | 9 of 10 | Newest score The newest came 2 years ago, in May 2024, for Gemini 1.5 Pro. | |
| Fund Smart Routing Agent | Professional work | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| Fundus-MMBench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| FutureOmni | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for MiMo-V2-Omni. | |
| FutureX | Math and science | 1 | 8 of 9 | Newest score The newest came 16 months ago, in July 2025, for Grok 4. | |
| FVEval | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-70B. | |
| G2 Voice AI Agent Platform Evaluations | Agents and tools | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Gaia2-CLI | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| GAIK LA-murre dialect test | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Gemini 3.5 Flash. | |
| GAIK SKN dialect test | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Gemini 3.5 Flash. | |
| GalCoLA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Galileo Agent Leaderboard v2 | Agents and tools | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for Mistral Medium 3.1. | |
| Galileo LLM Hallucination Index RAG Special | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 2 years ago, in June 2024, for Claude 3.5 Sonnet. | |
| Game Arena 3 Check Chess (Generality) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Kimi K2. | |
| Game Arena Anti Chess (Generality) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Kimi K2. | |
| Game Arena Atomic Chess (Generality) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| Game Arena Chess Openings | Agents and tools | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| Game Arena Crazyhouse Chess (Generality) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Kimi K2. | |
| Game Arena Four-in-a-Row (Generality) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for Kimi K2 Thinking. | |
| Game Arena Horde Chess (Generality) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Kimi K2. | |
| Game Arena King of the Hill (Generality) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Kimi K2. | |
| Game Arena Racing Kings (Generality) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Kimi K2. | |
| Game Arena Werewolf | Agents and tools | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Mini. | |
| GameCraft-Bench | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| GameEngineBench | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| GameForgeBench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GameHorizon | Agents and tools | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Gaokao Math 2026 | Math and science | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| GAOKAO-Eval 2024 | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in June 2024, for Qwen2-72B. | |
| Gauntlet | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GAUSS | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| gBench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GDP.pdf | Vision, audio and robotics | 1 | 11 of 12 | Lab claims 1 of 7 pairs differs by more than 10 points, the largest for Muse Wonder (−24 pts). | |
| GDPevo | Professional work | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| GDPval v2 (OpenAI) | Professional work | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| GDPval-AA | Professional work | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| GDPval-AA v2 | Professional work | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| GDPval-MM | Professional work | 1 | 10 of 11 | Newest score The newest came 9 months ago, in February 2026, for MiniMax-M2.5. | |
| GDPWorks | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GEdit-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for Step1X-Edit v1.2. | |
| Gemini Robotics generalization benchmark | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Gemini Robotics On-Device data scaling evaluation | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GEN-Bench | Professional work | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. | |
| GenAI-Arena | Vision, audio and robotics | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| Gender Resolution in Speaker-Listener Dialogue Roles | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GENDER1PERSON | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| General Agent Benchmark (Moonshot) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| General refusals (xAI) | Safety and security | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| General-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 23 months ago, in December 2024, for Falcon3-7B-Instruct. | |
| General365 | Math and science | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5 397B-A17B. | |
| GenEval 2 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GenExam | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GenomeGym-BRCA1 | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GenomeQA | Safety and security | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| GenPuzzle | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Flash Image. | |
| GenScale | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Genspark Slides Benchmark | Professional work | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| GEOBench-VLM | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| GeoBiaset | Safety and security | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-8B-Instruct. | |
| GeoExpertQueries | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GeoGuessBench (panoramic) | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| GeoKTools | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GeoMap-Bench | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| GeoMind | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for π0.5. | |
| GeoMLAMA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GeoNatureAgent Benchmark | Agents and tools | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for GLM-5. | |
| German Argument Mining | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| German Traffic Sign Recognition Benchmark | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GermanQuAD | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Nemotron 3 Nano 4B. | |
| GermEval 2014 Named Entity Recognition | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| gg-bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for Claude 3.7 Sonnet. | |
| GIE-Bench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GigaAM Multilingual Internal In-the-Wild Speech Tests | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GIM | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7. | |
| GitTaskBench | Coding | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-14B. | |
| GJ.cool Vertical Book OCR Evaluation | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| GLES 2017 vote choice | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Glianorex | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for DeepSeek-V3. | |
| GLM-4 All Tools information seeking (Z.ai) | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in June 2024, for GLM-4. | |
| Global MMLU Lite | Knowledge and language | 1 | 10 of 11 | Lab claims 1 of 4 pairs differs by more than 10 points, the largest for Gemma 3 4B (−11 pts). | |
| Global MMLU Lite 2.0 | Knowledge and language | 1 | 4 of 5 | Importer The record links a board on kaggle.com that we don’t read yet. | |
| Global PIQA | Knowledge and language | 1 | 9 of 10 | Lab claims 1 of 5 pairs differs by more than 10 points, the largest for Qwen3.5-4B (+17 pts). | |
| GlobalMGSM | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GMAI-MMBench | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for InternVL-Chat-V1-5. | |
| GMAT Quantitative | Math and science | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| GMAT Verbal | Knowledge and language | 1 | 2 of 3 | Task access We don’t know whether the tasks are public. | |
| Goliath human mesh recovery | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GoodAI LTM Benchmark | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in April 2024, for Llama 3-70B. | |
| Google DeepMind scheming honeypot suite | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Google RE-Bench (GRB) | Coding | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Google Scanned Objects | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Google Universal Image Embedding Challenge | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Gorilla APIBench | Agents and tools | 1 | 4 of 5 | Newest score The newest came 2 years ago, in November 2024, for Llama 3.1-405B. | |
| GPQA Diamond | Math and science | 1 | 10 of 11 | Lab claims 7 of 129 pairs differ by more than 10 points, the largest for Magistral Small 1.2 (+22 pts). | |
| GPQA Extended | Math and science | 1 | 9 of 10 | Newest score The newest came 2 years ago, in March 2024, for Claude 3 Haiku. | |
| GPQA Main | Math and science | 1 | 10 of 11 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| GPT-3 Arithmetic | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GPT-3 Word Scrambling | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Grade Like a Human OS Dataset | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GRAFT | Safety and security | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Gram (Gauging Realistic Agentic Misbehavior) | Safety and security | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| GraphicDesignBench | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| GraspBench | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Gray Swan Reward Hacking Bench | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| GreekBarBench | Professional work | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for GPT-4.1. | |
| GreekBarRetrieval | Professional work | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for gpt-oss-120b. | |
| Green Job Description Entity Recognition | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Greptile AI Code Review Benchmark | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Gretel Synthetic PII Finance | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GroundCocoa | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in December 2023, for Gemini 1.0 Pro. | |
| Grounded Language Model Customer Evaluations (Contextual AI) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Group Binary Search (Riedl) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 16 months ago, in July 2025, for Qwen3-235B-A22B-Instruct. | |
| GroupAssistBench | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GSM-Infinite | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for DeepSeek-R1. | |
| GSM8K-V | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Kimi K2.6. | |
| GTA-2 | Agents and tools | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. | |
| GTBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 2 years ago, in January 2024, for GPT-3.5 Turbo. | |
| GTFS Semantics | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| GTZAN | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Guesswork (The Aggregate) | Math and science | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| GUI Knowledge Bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5 Chat. | |
| GUI Odyssey | Agents and tools | 1 | 9 of 10 | Newest score The newest came 2 years ago, in July 2024, for InternVL2-Pro. | |
| GUI-CEval | Agents and tools | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for MiMo-VL-7B-RL. | |
| GuiaCat | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for LFM2.5-1.2B-Thinking. | |
| H2R-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for LTX-2.3. | |
| HABIT-Bench | Knowledge and language | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Hack-Verifiable Terminal Bench | Safety and security | 1 | 7 of 8 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| HackerBench | Safety and security | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| HackSynth OverTheWire | Safety and security | 1 | 9 of 10 | Newest score The newest came 2 years ago, in August 2024, for Phi-3.5-MoE. | |
| HackSynth PicoCTF | Safety and security | 1 | 9 of 10 | Newest score The newest came 2 years ago, in August 2024, for Phi-3.5-MoE. | |
| HAGRID | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HALAS | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in October 2024, for Whisper large-v3-turbo. | |
| HalloMTBench | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HalluDial | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in May 2024, for GPT-4o. | |
| HalluLens | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HaluBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in March 2024, for Claude 3 Haiku. | |
| HaluEval | Knowledge and language | 1 | 4 of 5 | Newest score The newest came 8 months ago, in March 2026, for Nemotron 3 Nano 4B. | |
| HaluEvalAudio 1000 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Hands Visible | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Hangman Bench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HANS | Knowledge and language | 1 | 6 of 7 | Confidence Likely. Some facts rest on secondary sources. | |
| HappyWorld-Bench | Vision, audio and robotics | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| Hardest SAT Problems (Vellum) | Math and science | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| HarmEval | Safety and security | 1 | 7 of 8 | Newest score The newest came 2 years ago, in December 2023, for Mistral-7B-Instruct-v0.2. | |
| Harmlessness human feedback win rate (Anthropic) | Safety and security | 1 | 2 of 3 | Confidence Likely. Some facts rest on secondary sources. | |
| HARMOGEN-R | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HarnessAudit-Bench | Safety and security | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Kimi K2.6. | |
| HARP-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in October 2024, for π0. | |
| Harvest Rush Training Evaluation | Safety and security | 1 | 7 of 8 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-8B. | |
| HCSU-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| HD-EPIC VQA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for Qwen2.5-VL-7B. | |
| HDLBits | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HEAD-QA v2 | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-70B. | |
| HealBench-Eval | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| HealthBench Consensus Global Health | Professional work | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| HealthBench Consensus Urgent | Professional work | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| HealthBench Meta | Professional work | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for o3. | |
| HealthBench Professional | Professional work | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| HealthBench-Hallu | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Baichuan-M3-235B. | |
| Hebbia Financial Services Benchmark | Professional work | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| Hebbia Financial Services Benchmark (2025) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Hebrew LLM Chat Leaderboard | Knowledge and language | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Hedge Gym | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HellaSwag | Knowledge and language | 1 | 9 of 10 | Lab claims 1 of 5 pairs differs by more than 10 points, the largest for Gemma 3 27B (+17 pts). | |
| HelloBench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in September 2024, for o1-mini. | |
| HELM Arabic | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7. | |
| HELM Capabilities | Other | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| HELM Classic | Other | 1 | 8 of 9 | Newest score The newest came 3 years ago, in July 2023, for Llama 2-13B. | |
| HELM Lite | Other | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for DeepSeek-V3. | |
| HELM Safety | Safety and security | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| HELMET | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 16 months ago, in July 2025, for Falcon-H1-34B. | |
| HelpBench | Safety and security | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Hermes Index | Other | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| HiACC | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 10 months ago, in January 2026, for Qwen3-ASR-0.6B. | |
| Hidden agenda | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Hiding a secret password (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HiL-Bench | Coding | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| HireWatch | Professional work | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7. | |
| Histoires Morales | Safety and security | 1 | 8 of 9 | Newest score The newest came 3 years ago, in September 2023, for Mistral-7B-Instruct-v0.1. | |
| Historical Systems Bugs | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HLE Bio/Chem Gold | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HLE-Rolling | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HLE-Verified (text-only Full) | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| HLE-VL | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| HM3D-MFMON | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HMDB51 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HoloBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for Gemini 2.0 Flash. | |
| Homayouni and Fontes Flexible Job Shop with Transportation | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HomeLoop | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| HoneyBench v0.2-alpha | Safety and security | 1 | 3 of 4 | Task access We don’t know whether the tasks are public. | |
| Honor of Kings Game Configuration | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Hotter and Colder | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| How frontier models miss risk in routine enterprise work | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Huayu Awards AI Cockpit Evaluation 2026 | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Human Behavior Bench | Math and science | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Human feedback win rates (Anthropic) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HumanAgencyBench | Safety and security | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| HumanEval | Coding | 1 | 9 of 10 | Lab claims 1 of 18 pairs differs by more than 10 points, the largest for Mistral Large (−24 pts). | |
| HumanEval Infilling | Coding | 1 | 4 of 5 | Newest score The newest came 17 months ago, in June 2025, for Mercury Coder Small. | |
| HumanEval Multi | Coding | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for Kimi k1.5. | |
| HumanEval-X | Coding | 1 | 9 of 10 | Newest score The newest came 4 years ago, in April 2022, for GPT-NeoX-20B. | |
| HumanEvalPack | Coding | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Cohere Command A. | |
| Humanity's Last Exam | Knowledge and language | 1 | 10 of 11 | Known issues 1 finding disputed or fixed only in a copy: Wrong labels. | |
| Humor in AI: Cartoon Captioning | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in February 2024, for Claude 3 Opus. | |
| Hungarian National High-School Exam | Math and science | 1 | 10 of 11 | Newest score The newest came 2 years ago, in January 2024, for DeepSeek LLM 67B. | |
| Hx2Dx ARENA | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Hy-Nomas-Bench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HY3D-Bench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| HybridDeepResearch | Coding | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-122B-A10B. | |
| HyDRA-Bench | Coding | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Claude Opus 4.5. | |
| HydroLLM-Benchmark | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Hypersim | Vision, audio and robotics | 1 | 4 of 5 | Newest score The newest came 8 months ago, in March 2026, for Qwen3.5-9B. | |
| i2b2 2006 Smoking Challenge | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| i2b2 2008 Obesity Challenge | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| i2b2 2009 Medication Extraction Challenge | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| i2b2/UTHealth 2014 Identifying Risk Factors for Heart Disease over Time | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| IARPA Babel speech recognition | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| IberoBench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in April 2024, for Llama 3-8B. | |
| ICML 2025 Experts Benchmark | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Pro. | |
| ICNALE Edited Essays | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ICNALE Global Rating Archives | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| IdeaTaste | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| IEMOCAP | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| IFEvalCode | Coding | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-0.6B. | |
| IFMTBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Qwen 3.6 35B-A3B. | |
| IFScale | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| IG-Bench | Math and science | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| IHEval | Safety and security | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| IITP-Legal | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Pro. | |
| Image to Text Safety | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Image-to-WebDev Arena | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Image2Struct | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in October 2024, for Claude 3.5 Sonnet (October 2024). | |
| IMB | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in July 2024, for Gemma 2 2B IT. | |
| IMDb-PT | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ImgEdit-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Nova 2 Omni Preview. | |
| IMO-GradingBench | Math and science | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for Gemini 2.5 Deep Think. | |
| IMO-LeanProofBench | Math and science | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. | |
| Implicit Intelligence | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7. | |
| ImpossibleBench | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| ImpossibleMLEBench | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| IMProofBench | Math and science | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| In-house CTF suite (Phuong et al. 2024) | Safety and security | 1 | 9 of 10 | Newest score The newest came 16 months ago, in July 2025, for Gemini 2.5 Flash. | |
| INAS hypothesis span prediction | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| iNaturalist 2018 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| INCLUDE Indian Sign Language Recognition | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Incorrect behavior rate on disallowed and sensitive content | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Indic DiarBench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| Indic-KCC-Agri-Advisory-Benchmark | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. | |
| IndicContextEval | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| IndicGenBench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| IndicIFEval | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| IndicIPR-QA | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in September 2024, for Qwen2.5-7B. | |
| IndicParam | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Claude Haiku 4.5. | |
| IndoMMLU | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 23 months ago, in December 2024, for Qwen2.5-14B. | |
| IndustryBench | Professional work | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Qwen 3.6 Plus. | |
| IneqMath | Math and science | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for DeepSeek-V3.1. | |
| InferenceEval | Coding | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| InfiniteBench/En.MC | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in November 2024, for Llama 3.1-405B. | |
| InfiniteBench/En.QA | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in November 2024, for Llama 3.1-405B. | |
| InFoBench | Knowledge and language | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| InfraBench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| INJONGO | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 23 months ago, in December 2024, for Llama 3.3 70B Instruct. | |
| InnovationEval | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Input hallucinations (Anthropic) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| INQUIRE-Rerank | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| InsCtrl (Tencent) | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Insider Trading Demo | Safety and security | 1 | 8 of 9 | Newest score The newest came 3 years ago, in September 2023, for GPT-3.5 Turbo Instruct. | |
| Insight Consolidation v1 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Inspect India Evals | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Instinct Next Edit | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Instruction Following (safety evaluation) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Instruction Hierarchy Evaluation | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| InstructTTSEval | Vision, audio and robotics | 1 | 4 of 5 | Newest score The newest came 10 months ago, in January 2026, for Qwen3-TTS-12Hz-1.7B-CustomVoice. | |
| InstrumentalEval | Safety and security | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for o3-mini. | |
| Intentionally taking actions very rarely (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Inter-GPS (Geometry3K) | Vision, audio and robotics | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| InteractiveBench Logic | Math and science | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| InteractiveBench Math | Math and science | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| InteractiveBench Texas Hold’em | Agents and tools | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| InteractiveBench Trust Game | Agents and tools | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| InteractScience | Coding | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| Internal Finance Modeling (Moonshot) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Internal Financial Research (Moonshot) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| InternBootcamp | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for DeepSeek-R1. | |
| Introductory Sommelier (theory knowledge) | Professional work | 1 | 2 of 3 | Task access We don’t know whether the tasks are public. | |
| Inverse IFEval | Knowledge and language | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Inverse Scaling Prize Tasks | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 4 years ago, in April 2022, for PaLM (540B). | |
| Investment Banking Modeling Tasks (OpenAI) | Professional work | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Investor's Last Exam | Professional work | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Invoice Approval (Ali Siala) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| IOI v2 (Vals) | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| iOSWorld | Agents and tools | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Mini. | |
| IPhO 2025 | Math and science | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| IPTC-test | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| ironbench | Professional work | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-4B. | |
| IronClad | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Irregular FrontierCyber and Atomic Challenges | Safety and security | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| ISA-Bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Granite 4.1 8B. | |
| iSarcasmEval | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| IsiZulu Second Language Learner Speech Corpus | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| IslamicFaithQA | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for gpt-oss-20b. | |
| IslamTrust | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ITBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Claude Opus 4.5. | |
| ITBench-AA | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Itzi Roleplay Benchmark: Local Robustness | Knowledge and language | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| IVISTA Voice Interaction, revised 2023 protocol | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| IWSLT 2017 TED Translation | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| JA-VLM-Bench-In-the-Wild | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 2 years ago, in January 2024, for llava-v1.6-mistral-7b. | |
| JailbreakBench | Safety and security | 1 | 7 of 8 | Newest score The newest came 10 months ago, in January 2026, for Falcon H1R 7B. | |
| JamC-QA | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for Qwen2.5-72B. | |
| Japanese LLM Leaderboard (lm-evaluation-harness) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Japanese National Dental Examination 2025 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Japanese Tort-case Dataset | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Japanese-English Bilingual Corpus of Wikipedia’s Kyoto Articles | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| JAQKET | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Jarvis Bench | Vision, audio and robotics | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Jawaher | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| JCoLA | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in January 2024, for GPT-3.5 Turbo. | |
| JEMHopQA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| JETHICS | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in November 2024, for GPT-4o. | |
| JFinQA | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| JGPQA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| JHumanEval | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Jiminy Cricket | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| JMultiWOZ-TC | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| JorGPT Student Answer Grading | Professional work | 1 | 7 of 8 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash-Lite. | |
| JSeM | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| JSPLIB | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Judge Arena (Atla) | Other | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| JudgeBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Llama-3.1-Nemotron-Ultra-253B-v1. | |
| Judgemark v4 | Knowledge and language | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| JudgerBench V2 | Other | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-235B-A22B. | |
| K-QA | Professional work | 1 | 9 of 10 | Newest score The newest came 3 years ago, in June 2023, for GPT-4. | |
| K-Viscuit | Vision, audio and robotics | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| K12-Bench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 31B IT. | |
| K12-PEBench | Professional work | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for Qwen2.5-VL-72B. | |
| K12Vista | Professional work | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for Qwen2.5-VL-32B. | |
| K9-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-32B-Thinking. | |
| Kaeru Flip (paprika3d) | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Kaggle Chess Suite | Agents and tools | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| Kaggle's Task Tuesday Benchmark | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| KairosQA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Kaleidoscope | Knowledge and language | 1 | 10 of 11 | Newest score The newest came 18 months ago, in May 2025, for Aya Vision 32B. | |
| KARLBench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Claude Sonnet 4.6. | |
| Kazakh Speech Corpus 2 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| KBank-MMLU (Upstage) | Professional work | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Keeping secrets when pressed in simulated scenarios (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| KernelBench | Coding | 1 | 5 of 6 | Importer The record links a board on scalingintelligence.stanford.edu that we don’t read yet. | |
| KernelBench-Verified | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| KernelBenchFiltered | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| KernelBenchInternal | Coding | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| KeSpeech | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| KeyFrame-Compass | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for LTX-2.3. | |
| Keystone-Eval | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Khmer Handwritten Dataset (4.2k) | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| KilometerVision | Vision, audio and robotics | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| Kimi Code Bench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Kimi Code Bench 3.0 Long-horizon | Coding | 1 | 11 of 12 | Confidence Likely. Some facts rest on secondary sources. | |
| Kimi Promptfoo red-team safety evaluation | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Kimi vulnerability discovery (Tier 1) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Kimi-ASR Internal Testset | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Kimi-ASR Internal Testset Subset 1 | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Kimi-Audio-7B-Instruct. | |
| Kimi-ASR Internal Testset Subset 2 | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Kimi-Audio-7B-Instruct. | |
| Kinetics-700 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Kinetics-700-2020 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for Qwen3-VL-Embedding-2B. | |
| Kitab | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 19 months ago, in April 2025, for Phi-4-Reasoning-plus. | |
| KMMLU | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for Solar Open 100B. | |
| KMMLU-Pro 2025 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| KMMLU-Pro 2026 | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 16 months ago, in July 2025, for EXAONE 4.0 32B. | |
| KMMLU-Redux | Knowledge and language | 1 | 10 of 11 | Newest score The newest came 16 months ago, in July 2025, for EXAONE 4.0 32B. | |
| KMMMU | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for EXAONE 4.5 33B. | |
| KnowledgeGraphEval 2026 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| KnowMe-Bench | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| Known-Unknown Questions | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 3 years ago, in February 2023, for LLaMA-13B. | |
| KnowU-Bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-9B. | |
| Ko-AIME 2025 (Upstage) | Math and science | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| KoBALT | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for EXAONE 4.5 33B. | |
| KoBEST | Knowledge and language | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| KoCommonGEN v2 | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-8B. | |
| KOR-Bench | Math and science | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Seed1.8. | |
| KORGym | Agents and tools | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-32B. | |
| KorSET | Safety and security | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for Gemma 3 12B. | |
| KPI-EDGAR | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| KradleBench | Agents and tools | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| KradleVerse | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| KRIS-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Step Image Edit 2. | |
| KurdiSent | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Kyoto University Commonsense Inference Dataset | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Kyoto University Wikipedia Annotated Corpus | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| L-Eval | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in February 2024, for Qwen1.5-72B. | |
| L2-Bench | Professional work | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7. | |
| LAB: Contracts | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LAB: Diligence | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LABBench2 CloningQA | Math and science | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| LABBench2 FigQA2 | Safety and security | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| LABBench2 FigQA2 (image) | Math and science | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| LABBench2 FigQA2 (PDF) | Math and science | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| LABBench2 ProtocolQA2 | Safety and security | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| LABBench2 SeqQA2 | Safety and security | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| Labelbox Agentic Search | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Pro. | |
| Labelbox Deep Research Agents | Agents and tools | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Pro. | |
| Labelbox Image Generation (original) | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Labelbox Image Generation v2 | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for GPT-4.1. | |
| Labelbox Multimodal Reasoning | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| Labelbox Speech Generation (original) | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Labelbox Video Generation (original) | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LAMBADA Multilingual (Stable LM 2) | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in February 2024, for Gemma 2B. | |
| LAMBADA OpenAI Multilingual | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Lamina Product Bench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LaminaBench v1 | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LangChain Benchmarks | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Last Translation Benchmark | Knowledge and language | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| LawBench | Professional work | 1 | 8 of 9 | Newest score The newest came 3 years ago, in September 2023, for Qwen-7B. | |
| LBM Eval | Vision, audio and robotics | 1 | 4 of 5 | Newest score The newest came 7 months ago, in April 2026, for Foundry-Qwen3VLA-2.1B. | |
| LCBench v0.1 | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Learn2Play Bench | Agents and tools | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Leeds Sports Pose Extended | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Legal World sample | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LegalCiteBench | Professional work | 1 | 8 of 9 | Newest score The newest came 13 months ago, in October 2025, for Claude Haiku 4.5. | |
| LegalRewardBench | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| LegalSupport | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LEGO-Puzzles | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-32B-Instruct. | |
| Leipzig Benchmark | Math and science | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| LengthCtrl (Tencent) | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Letta Leaderboard | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| LEVIR-CDC | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LEXam | Professional work | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. | |
| LEXam-hard | Professional work | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| LexGenius | Professional work | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-4B. | |
| LHRS-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in February 2024, for Claude 3 Opus. | |
| LIBERO-MAX | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for MolmoAct 2. | |
| LIBERO-Para | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for π0.5. | |
| LIBERO-Plus | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for OpenVLA-OFT. | |
| LIBERO-PRO | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for π0.5. | |
| LIBRA (Russian Long Context) | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-0.8B. | |
| LibriCount | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 3 years ago, in November 2023, for Whisper large-v3. | |
| Life-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for Gemma 3 12B. | |
| LifeChoice | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 23 months ago, in December 2024, for o1. | |
| LifeSci Diamond | Professional work | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| LifeSciBench | Safety and security | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| LifeVal | Agents and tools | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Lights Out 3×3 | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Linear Algebra Done Right (Axler Exercises) | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LingOly | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in May 2024, for GPT-4o. | |
| LingOly-TOO | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for DeepSeek-V3.1-Terminus. | |
| LinguaSafe | Safety and security | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Apertus 70B. | |
| Live Gaming Benchmark | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for LiveCC-7B-Instruct. | |
| LiveDRBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 21 months ago, in February 2025, for OpenAI deep research. | |
| LiveMath | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LiveMathBench | Math and science | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for o3-mini. | |
| LiveNewsBench Release 2 | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for Kimi K2.5. | |
| LiveOCRVQA | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| LiveSearch | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| LiveSports-3K | Vision, audio and robotics | 1 | 4 of 5 | Task access We don’t know whether the tasks are public. | |
| LiveTradeBench | Professional work | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| Llama 2-Chat helpfulness human evaluation (Meta) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Llama 2-Chat safety human evaluation (Meta) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Llama 3 pairwise human evaluation (Meta) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Llama internal safety benchmarks (Meta) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Llama verbatim memorization probe (Meta) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LLaVA-Bench (In-the-Wild) | Vision, audio and robotics | 1 | 6 of 7 | Confidence Likely. Some facts rest on secondary sources. | |
| LLM Confabulation Leaderboard | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| LLM Deceptiveness and Gullibility | Agents and tools | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| LLM Divergent Thinking Creativity Benchmark | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| LLM Game Benchmark | Agents and tools | 1 | 8 of 9 | Newest score The newest came 2 years ago, in June 2024, for Claude 3.5 Sonnet. | |
| LLM Games | Agents and tools | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| LLM Persuasion Benchmark | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for MiniMax-M2.7. | |
| LLM Public Goods Game | Agents and tools | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for DeepSeek-R1. | |
| LLM Round-Trip Translation Benchmark | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Kimi K2. | |
| LLM SoccerArena Bundesliga 2026/27 | Math and science | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| LLM SoccerArena La Liga 2026/27 | Math and science | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| LLM SoccerArena Premier League 2026/27 | Math and science | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| LLM Thematic Generalization Benchmark v2 | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7. | |
| LLM Training Puzzles | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LLM Uncertainty Bench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in November 2023, for Qwen-1_8B. | |
| LLM Writer Negative Style | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 19 months ago, in April 2025, for GPT-4.1 mini. | |
| LLM-AggreFact | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 23 months ago, in December 2024, for Llama 3.3 70B Instruct. | |
| LLM-Coordination | Agents and tools | 1 | 8 of 9 | Newest score The newest came 2 years ago, in December 2023, for Mixtral 8x7B. | |
| llm-jp-eval | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LLM-NERRE materials extraction | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LLM-Perf Leaderboard | Other | 1 | 8 of 9 | Newest score The newest came 2 years ago, in June 2024, for Qwen2-0.5B. | |
| LLM-SRBench | Math and science | 1 | 8 of 9 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-8B. | |
| LLM-WikiRace | Agents and tools | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| LLMBar | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LLMEval-Med | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LLMzSzŁ | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-1.7B. | |
| LMAct | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LMArena (Elo) | Knowledge and language | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| LMArena Image Edit Arena | Vision, audio and robotics | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| LMArena Text-to-Image Arena | Vision, audio and robotics | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| LMArena Video Edit Arena | Vision, audio and robotics | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| LMentry | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 4 years ago, in March 2022, for davinci-002. | |
| lmgame-Bench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| Local LLM Arkanoid Test | Coding | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| Local LLM Tetris Test | Coding | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| LocalSearchBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. | |
| LogiQA | Knowledge and language | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| LOKI | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in August 2024, for Qwen2-VL-72B-Instruct. | |
| Long Code Arena | Coding | 1 | 8 of 9 | Newest score The newest came 23 months ago, in December 2024, for o1. | |
| Long-horizon jailbreaks (xAI) | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| LongAudioSpan | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| LongAV-Compass | Vision, audio and robotics | 1 | 6 of 7 | Confidence Likely. Some facts rest on secondary sources. | |
| LongBanana | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Flash Image. | |
| LongBench English | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 16 months ago, in July 2025, for EXAONE 4.0 32B. | |
| LongBench Pro | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Ministral 3 3B Instruct. | |
| LongCodeBench | Knowledge and language | 1 | 4 of 5 | Newest score The newest came 11 months ago, in December 2025, for Nova 2.0 Pro Preview. | |
| LongCoT | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-27B. | |
| LongDocURL | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Seed2.0 Pro. | |
| LongExtractionBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| LongFact-Concepts | Knowledge and language | 1 | 3 of 4 | Newest score The newest came 9 months ago, in February 2026, for Seed2.0 Pro. | |
| LongFact-Objects | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Seed2.0 Pro. | |
| LongHealth | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in December 2023, for Mistral-7B-Instruct-v0.2. | |
| LongJud-Bench | Professional work | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. | |
| LongLeader | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| LongProc | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in October 2024, for Claude 3.5 Sonnet (October 2024). | |
| LongShOTBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Qwen3.6 27B. | |
| LoopsBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| LP-TTS-Bench | Vision, audio and robotics | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| LT-History | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for Llama 3.2 3B. | |
| ltzGLUE | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5 mini. | |
| LV-Eval | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| LVSum | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| m-ArenaHard v2.0 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| m-ArenaHard v2.1 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| M-RewardBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in October 2024, for Aya Expanse 32B. | |
| M3-Bench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| M3Exam | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| M3GIA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for STEP3-VL-10B. | |
| M4Singer | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MA-Bench | Other | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| MaCBench | Math and science | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| MacroLens | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MAD-Bench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MAD-Eval | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MADE 1.0 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MadinahQA | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Falcon-Arabic-7B. | |
| MAFAND-MT | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MafiaBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 21 months ago, in February 2025, for Claude 3.7 Sonnet. | |
| mage-bench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Mini. | |
| Magentic Marketplace | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MagnaTagATune | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MAI internal jailbreak evaluation suite | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| MAI internal safety and over-refusal evaluation | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| MAI Thinking 1 blind human preference evaluation | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MakeMePay | Safety and security | 1 | 5 of 6 | Newest score The newest came 21 months ago, in February 2025, for GPT-4.5. | |
| MakeMeSay | Safety and security | 1 | 5 of 6 | Newest score The newest came 7 months ago, in April 2026, for Grok 4.20. | |
| Malaysian Sign Language Medical Dataset | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Malicious agentic influence campaigns (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Malicious agentic influence campaigns, 34-point rubric (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Málrómur | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Malt Freelancer Project Matching | Professional work | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Qwen3-Embedding-0.6B. | |
| Maltese News Categories | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MAMUBench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Management Consulting Tasks (OpenAI) | Professional work | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| Manager Coercion Benchmark | Safety and security | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| ManipArena | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MANTA (Animal Welfare) | Safety and security | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Grok 4.3. | |
| Mantis-Eval | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for LongCat-Flash-Omni. | |
| Manually-Reviewed Agentic Labeled Transcripts | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Many Labs 2 text-based replications | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Many-shot jailbreaking evaluation (Anthropic) | Safety and security | 1 | 2 of 3 | Confidence Likely. Some facts rest on secondary sources. | |
| MapAnything dense multi-view reconstruction | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MAPP (Multi-Agent Path Planning) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for GPT-4.1. | |
| MAPS Piano Transcription | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MarathonDocBench | Vision, audio and robotics | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| MarineEval | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for InternVL3-38B. | |
| MarineEVT | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| Markov Blanket Discovery (shinpaku) | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MARS-Bench | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| MasakhaNEWS | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MasakhaPOS | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in August 2024, for GPT-4o. | |
| Massively Multilingual Speech labeled corpus | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Mastermind (Prime Intellect) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for GLM-4.6. | |
| Mastermind Information Gain | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MastermindEval | Agents and tools | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for DeepSeek-R1-Distill-Qwen-1.5B. | |
| MatBERT gold nanoparticle entity recognition | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MatCha | Math and science | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for InternVL3-38B. | |
| Materials Science Procedural Text Corpus | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MATH Level 5 | Math and science | 1 | 5 of 6 | Importer The record links a board on epoch.ai that we don’t read yet. | |
| Math-VR | Vision, audio and robotics | 1 | 10 of 11 | Newest score The newest came 12 months ago, in November 2025, for Qwen3-VL-235B-A22B. | |
| Math23K | Math and science | 1 | 2 of 3 | Confidence Likely. Some facts rest on secondary sources. | |
| MathBench | Math and science | 1 | 9 of 10 | Newest score The newest came 2 years ago, in June 2024, for Qwen2-72B. | |
| Mathematical Problem Repository for AlphaEvolve | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MathIF | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-0.6B. | |
| MathOlympiadBench | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MathQA | Math and science | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Apertus 70B. | |
| MathVision-Wildphoto | Vision, audio and robotics | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| MatSciBench | Math and science | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| Matter-of-Fact | Math and science | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for o4-mini. | |
| MatToolBench | Math and science | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| MatTools | Math and science | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for Gemini 2.0 Flash. | |
| MAVERIX | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Nova 2 Omni Preview. | |
| MAWPS | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MaXM | Vision, audio and robotics | 1 | 4 of 5 | Newest score The newest came 18 months ago, in May 2025, for Aya Vision 32B. | |
| Maydan HF RL Sampler | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MBE (Multistate Bar Examination) | Professional work | 1 | 2 of 3 | Task access We don’t know whether the tasks are public. | |
| MBPP+ | Coding | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| MBTI Post Classification (Kunumi) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| McEval | Coding | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Hunyuan-A13B-Instruct. | |
| MCP-Bench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| MCP-SafetyBench | Safety and security | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| MCP-SIM | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MCP-Universe | Agents and tools | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. | |
| MCPTox | Safety and security | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-14B. | |
| MCS-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| MdCoEditBench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MDMR-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Flash Image. | |
| mDolly | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MDPBench | Vision, audio and robotics | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| MeasureBench Real-world | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| MeasureBench Synthetic | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| Med-HALT | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in July 2023, for Llama 2-13B. | |
| MedAgentBench | Professional work | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for Gemini 2.0 Flash. | |
| MedAlign | Professional work | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| MedBench 1.0 (OpenCompass) | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Medbullets | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for Llama 3-70B. | |
| MedChemBench (OpenAI) | Safety and security | 1 | 11 of 12 | Lab claims 3 of 4 pairs differ by more than 10 points, the largest for GPT-6 Astra (−35 pts). | |
| MedConceptsQA | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in January 2024, for GPT-3.5 Turbo. | |
| MEDDOPROF | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MEDEC | Professional work | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for Gemini 2.0 Flash. | |
| MedGemma lab report conversion | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MedHallu | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in July 2024, for Gemma 2 2B IT. | |
| MedHEB | Professional work | 1 | 8 of 9 | Newest score The newest came 10 months ago, in January 2026, for Qwen3-VL-Embedding-2B. | |
| MedHELM | Professional work | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| Media Frames Corpus | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MediaPerf | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Nemotron 3 Nano Omni 30B-A3B. | |
| Medical Chronology LLM Benchmark | Professional work | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Mini. | |
| Medical Paper-Blog Dataset | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Medical Prescription Selection (lm-evaluation-harness) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Medical Speech-to-Text Benchmark | Vision, audio and robotics | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| MedicationQA | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MEDIQA 2019 Question Answering | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MEDIQA-SYNUR | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MedLFQA | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in February 2024, for Claude 3 Sonnet. | |
| MedMCQA | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in August 2024, for GPT-4o. | |
| MedMentions | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MedPhysBench Knowledge Questions | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MedPT | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MedQA Bias Evaluation | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| MedQA Mainland China | Professional work | 1 | 4 of 5 | Newest score The newest came 2 years ago, in August 2024, for GPT-4o. | |
| MedQA-Followup | Professional work | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Claude Opus 4.5. | |
| MedSafetyBench | Safety and security | 1 | 8 of 9 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-70B. | |
| MedText | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MedVidQA 2023 Video Corpus Visual Answer Localization | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MedVidQA 2024 Query-Focused Instructional Step Captioning | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MedVidQA 2024 Video Corpus Visual Answer Localization | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MEEN 361 Conceptual Question Grading | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Meeseeks | Knowledge and language | 1 | 4 of 5 | Importer The record links a board on github.com that we don’t read yet. | |
| MEGA-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for Qwen2.5-VL-72B. | |
| MegaGem | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Megaminx solver | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MegaOath Forecasting Survey | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MELA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Mem-Gallery | Vision, audio and robotics | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| Mementos | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 3 years ago, in November 2023, for GPT-4V. | |
| MemoryAgentBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5 mini. | |
| Menagerie: A Dataset of Graded Programming Assignments | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MentalBench (Psychiatric Diagnosis) | Professional work | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for GPT-5.1. | |
| MentalBench-100k | Professional work | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-4B. | |
| MentalChat16K Counseling Evaluation | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in December 2023, for Mistral-7B-Instruct-v0.2. | |
| MentalHealthBench | Safety and security | 1 | 11 of 12 | Confidence Likely. Some facts rest on secondary sources. | |
| MERA | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. | |
| Mercury | Coding | 1 | 8 of 9 | Newest score The newest came 2 years ago, in April 2024, for CodeQwen1.5-7B. | |
| MermaidSeqBench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MESA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GR00T N1.6. | |
| MET-Meme | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Meta BioKnowledge Proxy | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Meta chemical and biological high-risk bottleneck evaluations | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| metabench | Other | 1 | 8 of 9 | Newest score The newest came 2 years ago, in May 2024, for Yi-1.5-9B-Chat. | |
| metal-arm-harness | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| METR Faithfulness and Monitorability Experiments | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| METR Time Horizon | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| METR Time Horizon suite | Coding | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for Claude Opus 4.5. | |
| mFollowIR | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 3 years ago, in September 2023, for Mistral-7B-Instruct-v0.1. | |
| MHPP | Coding | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Seed-Coder-8B-Instruct. | |
| MiJaBench | Safety and security | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-1.7B. | |
| Miklós Schweitzer 2025 | Math and science | 1 | 6 of 7 | Confidence Likely. Some facts rest on secondary sources. | |
| MiMo Code Bench | Coding | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| MiMo Cyber Bench | Safety and security | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| MiMo Visual Coding | Coding | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| Mind the Gap | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Gemini 3.5 Flash. | |
| Mind2Web 2 | Agents and tools | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for o3. | |
| MindGames | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MindWorldBench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MineAmongUs | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Qwen3.6 27B. | |
| Minecraft Bed Wars (YC Labs) | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MineDojo | Agents and tools | 1 | 4 of 5 | Newest score The newest came 9 months ago, in February 2026, for Seed2.0 Pro. | |
| MINERVA-Cultural | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| Minerva-Ego | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| Ming-Freeform-Audio-Edit | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Mini Amusement Parks | Agents and tools | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Claude Sonnet 4.5. | |
| miniCTX-v2 | Math and science | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| MiniStS | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MiniWoB++ | Agents and tools | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| MIR-1K | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MIRAGE-MMMT | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Llama 4 Maverick. | |
| MIRAGE-MMST | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for GPT-4.1. | |
| MIRB | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for InternVL3.5-241B-A28B. | |
| MirrorCode | Coding | 1 | 7 of 8 | Importer The record links a board on epoch.ai that we don’t read yet. | |
| MisconceptDialogue | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Misguided Attention | Math and science | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Gemma 3n E4B. | |
| Missing-context hallucinations (Anthropic) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Mistral self-reflection moderation set | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Mixer 6 Speech | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MixEval | Knowledge and language | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| MIZAN: A Persian LLM Leaderboard | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| MJ-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for InternVL-Chat-V1-5. | |
| MK-Bench | Professional work | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. | |
| MLAgentBench | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in December 2023, for Gemini 1.0 Pro. | |
| MLCommons AI Safety Benchmark v0.5 | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MLCommons AILuminate Jailbreak | Safety and security | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| MLCR-AA | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MLE-bench 30 | Coding | 1 | 5 of 6 | Confidence Likely. Some facts rest on secondary sources. | |
| MLGym-Bench | Coding | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Pro. | |
| MLogiQA | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Apertus 70B. | |
| MLQA | Knowledge and language | 1 | 6 of 7 | Confidence Likely. Some facts rest on secondary sources. | |
| MLRC-Bench | Coding | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| MM-AlignBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in November 2024, for InternVL2-8B-MPO. | |
| MM-BrowseComp | Agents and tools | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| MM-ClawBench | Agents and tools | 1 | 10 of 11 | Newest score The newest came 6 months ago, in May 2026, for MiniMax-M2.5. | |
| MM-IFEval | Knowledge and language | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| MM-MT-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 10 months ago, in January 2026, for STEP3-VL-10B. | |
| MM-NIAH | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for Idefics2-8B. | |
| MM-OCEAN | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MM-ToolSandbox | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MM-Vet | Vision, audio and robotics | 1 | 10 of 11 | Newest score The newest came 8 months ago, in March 2026, for Reka Edge. | |
| MMAE | Vision, audio and robotics | 1 | 6 of 7 | Confidence Likely. Some facts rest on secondary sources. | |
| MMAU v05.15.25 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Nova 2 Omni Preview. | |
| MMBench-EN v1.1 | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| MMCircuitEval | Professional work | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Reka Edge. | |
| MMDU | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in April 2024, for Idefics2-8B. | |
| MME | Vision, audio and robotics | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| MME-RealWorld | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Nemotron Nano 12B v2 VL. | |
| MME-Reasoning | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 16 months ago, in July 2025, for Skywork-R1V3-38B. | |
| MME-VideoOCR | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Qwen3.5-Omni-Plus. | |
| MMEB-V3 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for Qwen3-VL-Embedding-2B. | |
| MMIU | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for InternVL3.5-241B-A28B. | |
| MMK12 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| MMLU | Knowledge and language | 1 | 8 of 9 | Known issues 1 finding disputed or fixed only in a copy: Wrong labels and unclear tasks. | |
| MMLU College Medicine | Professional work | 1 | 4 of 5 | Newest score The newest came 2 years ago, in August 2024, for GPT-4o. | |
| MMLU Early Training | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MMLU Medical | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in May 2024, for MAmmoTH2-8B-Plus. | |
| MMLU translated with Azure Translate | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MMLU-CF | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 23 months ago, in December 2024, for Phi-4. | |
| MMLU-ProX Lite | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-235B-A22B. | |
| MMLU-Redux 2.0 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MMLU-Redux 2.0 Spanish | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MMLU-SR | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in July 2024, for GPT-4o mini. | |
| MMLU-stem | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 16 months ago, in July 2025, for Falcon-H1-34B. | |
| MMMLU-Lite | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MMT-Bench | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| MMTU | Coding | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| MMVP | Vision, audio and robotics | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| Mobile Actions | Agents and tools | 1 | 2 of 3 | Confidence Likely. Some facts rest on secondary sources. | |
| Mobile-Eval | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MobileAgentBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 2 years ago, in December 2023, for CogAgent. | |
| MobileGym | Agents and tools | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Qwen 3.6 Plus. | |
| Model Spec Evals | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MoI-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Flash Image. | |
| Molecular Biology Capabilities Test | Safety and security | 1 | 11 of 12 | Task access We don’t know whether the tasks are public. | |
| MolmoAct2 sim_eval | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MolmoSpaces | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for MolmoAct 2. | |
| MOMENT Eval | Professional work | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Monitor Evasion Capability Under Prompted Sandbagging | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Monsoon ASR | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MooseBench | Coding | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. | |
| Moral Reasoning under Uncertainty | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Moral Stories | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MoReBench | Safety and security | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash-Lite. | |
| MoReBench-Theory | Safety and security | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash-Lite. | |
| Mosaic Eval Gauntlet Original | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Mosaic Eval Gauntlet v0.3.0 | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 3 years ago, in September 2023, for Mistral-7B-Instruct-v0.1. | |
| MotoVerse | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-2B-Instruct. | |
| MPDocBench-Parse | Vision, audio and robotics | 1 | 4 of 5 | Task access We don’t know whether the tasks are public. | |
| MPIE-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for BAGEL-7B-MoT. | |
| MRCR (Michelangelo, original) | Knowledge and language | 1 | 3 of 4 | Task access We don’t know whether the tasks are public. | |
| MRI Dataset 1 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MSEarth | Vision, audio and robotics | 1 | 10 of 11 | Newest score The newest came 8 months ago, in March 2026, for Intern-S1-Pro. | |
| MSR-VTT | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MSR-VTT retrieval | Vision, audio and robotics | 1 | 3 of 4 | Newest score The newest came 6 months ago, in May 2026, for Gemini Embedding 2. | |
| MSVD | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MT-AIME2024 | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Qwen3-1.7B. | |
| MT-Bench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Granite 4.1 30B. | |
| MT-Bench-101 | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for InternLM3-8B-Instruct. | |
| MTEB-BR | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for Gemini Embedding 2. | |
| MTG-Jamendo | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MTGBench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MTS-Dialog | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MUCS 2021 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MulSeT | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for Qwen2.5-VL-7B. | |
| Multi-Crit | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-8B-Instruct. | |
| Multi-SWE-bench | Coding | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for MiniMax-M2.5. | |
| Multi-SWE-bench flash | Coding | 1 | 10 of 11 | Newest score The newest came 16 months ago, in July 2025, for Qwen3-Coder-480B-A35B. | |
| Multi-SWE-bench mini | Coding | 1 | 10 of 11 | Newest score The newest came 16 months ago, in July 2025, for Qwen3-Coder-480B-A35B. | |
| Multi-Turn Human Jailbreaks | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Multi-Turn Sudoku (m8ngotree) | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Multi-turn testing (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| MultiChallenge | Knowledge and language | 1 | 11 of 12 | Lab claims 1 of 7 pairs differs by more than 10 points, the largest for Qwen3.5-4B (+24 pts). | |
| MultiClinAI | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MultiClinSum-2 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MultiCochrane | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MultiFinBen | Professional work | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Llama 4 Scout. | |
| MultiFinGround | Professional work | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| MultiGEC 2025 | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-8B. | |
| MultiHop-RAG | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in November 2023, for Claude 2.1. | |
| Multilingual AIME 2024 (Mistral) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Multilingual Conversational Speech Language Model Challenge | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Multilingual Entity Linking of Skills | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Multilingual Exams | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for Qwen2.5-72B. | |
| Multilingual HateCheck | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Multilingual human evaluation | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in July 2024, for Qwen2-72B. | |
| Multilingual IFEval | Knowledge and language | 1 | 2 of 3 | Task access We don’t know whether the tasks are public. | |
| Multilingual LibriSpeech | Vision, audio and robotics | 1 | 3 of 4 | Newest score The newest came 11 months ago, in December 2025, for Nova 2 Omni Preview. | |
| Multilingual performance on translated expert-domain tasks | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Multilingual Safety | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| MultiMedia-TerminalBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-122B-A10B. | |
| MultiMedQA | Professional work | 1 | 9 of 10 | Newest score The newest came 4 years ago, in April 2022, for PaLM (540B). | |
| Multimodal RewardBench 2 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| MultipanelVQA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in January 2024, for Gemini 1.0 Pro Vision. | |
| MUSDB18 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MUSE | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| MUSE (CAD assemblies) | Professional work | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| Muse-targeted cyber misuse and violation red-teaming | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Music Arena | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Stable Audio 3 Medium. | |
| Music Arena Vocal | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for Lyria 3 Clip. | |
| Music Understanding Benchmark | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MusICA-MetaBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Gemini 3.1 Flash-Lite. | |
| Musical Expectation Benchmark | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MusicTheoryBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Mini. | |
| MusiXQA | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in May 2024, for Phi-3-vision-128k-instruct. | |
| MUStARD | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MuTox | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MVPBench | Vision, audio and robotics | 1 | 4 of 5 | Importer The record links a board on huggingface.co that we don’t read yet. | |
| MyCareersFuture Skill Retrieval | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| MyClaw Bench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Mystery Game Puzzles | Agents and tools | 1 | 7 of 8 | Importer The record links a board on epoch.ai that we don’t read yet. | |
| N-1 Contingency Dispatch | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| n2c2 2018 Adverse Drug Events and Medication Extraction | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| n2c2 2022 Contextualized Medication Event Extraction | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| n2c2 2022 Progress Note Understanding: Assessment and Plan Reasoning | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NaijaProverb-Bench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| NaijaSenti | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Qwen3-Embedding-8B. | |
| NanoGPT Speedrun Frontier | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| NanoGPT-Bench | Coding | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| NAPISUL-II Progress Test 2023 | Professional work | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| NARRA-Gym | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| NarrativeQA | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 21 months ago, in February 2025, for Gemini 2.0 Flash. | |
| NarrativeTrack | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NarrLV | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 21 months ago, in February 2025, for Wan2.1-T2V-14B. | |
| NASA SDE IR Benchmark v5 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NASA-IR | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Natural Questions | Knowledge and language | 1 | 5 of 6 | Importer The record links a board on ai.google.com that we don’t read yet. | |
| Natural Questions in Icelandic | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NatureBench | Math and science | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| NB Tale Module 1 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NB Tale Module 3 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NBABench (Thierry Damiba) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NCHLT isiZulu Speech Corpus | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NCP-Bench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 10 months ago, in January 2026, for Kimi K2.5. | |
| NeedleBench V2 | Knowledge and language | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| Nego Eval | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Nejumi LLM Leaderboard 2 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NEPABench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NEPAQuAD | Professional work | 1 | 9 of 10 | Newest score The newest came 3 years ago, in September 2023, for Mistral-7B-Instruct-v0.1. | |
| NepTTS-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for Gemini 2.5 Flash TTS. | |
| NEREL-bench | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NEST | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Gemini 3.1 Flash-Lite. | |
| Neural Activation Reading for Collusion Benchmark | Safety and security | 1 | 7 of 8 | Newest score The newest came 15 months ago, in August 2025, for gpt-oss-20b. | |
| neuralCAD-Edit | Professional work | 1 | 7 of 8 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| NeurIPS 2025 Experts Benchmark | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| Neuro Reasoning (dwin2) | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NeuroDiscoveryBench | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NeuroscienceBench | Math and science | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| NEVIS’22 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| New Synonym Benchmark | Knowledge and language | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| New Yorker Caption Contest Humor Understanding | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 years ago, in October 2019, for T5-11B. | |
| News With GPT Instructions | Professional work | 1 | 7 of 8 | Newest score The newest came 3 years ago, in July 2023, for Llama 2-13B. | |
| NewsBench (Chinese journalism) | Professional work | 1 | 8 of 9 | Newest score The newest came 3 years ago, in November 2023, for GPT-4 Turbo. | |
| NewsBench (Forum AI) | Knowledge and language | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Nexus | Agents and tools | 1 | 9 of 10 | Newest score The newest came 2 years ago, in November 2024, for Llama 3.1-405B. | |
| NFCorpus | Professional work | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for Qwen3-Embedding-0.6B. | |
| NIA Sovereign AI Benchmark Evaluation, Phase 2 | Other | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NICTA-PIBOSO | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NIILC Question Answering Dataset | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Nim compiler repair benchmark | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Nim Hard | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NitiBench | Professional work | 1 | 7 of 8 | Newest score The newest came 2 years ago, in September 2024, for Gemini 1.5 Pro. | |
| Nodal Pricing LMP | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NoLiMa-Hard | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Gemini 2.5 Flash. | |
| Noor | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NorEval | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash-Base. | |
| NormAd-Eti | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in February 2024, for OLMo-7B-Instruct. | |
| norregel | Professional work | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| NorSumm | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 3 years ago, in September 2023, for Mistral 7B v0.1. | |
| Norwegian Comma Benchmark | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Norwegian Review Corpus | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NOTA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in May 2024, for MiniCPM-Llama3-V 2.5. | |
| NoticIA | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in February 2024, for Qwen1.5-14B-Chat. | |
| NOTSOFAR-1 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NotSoTiny 25-12 | Professional work | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 31B IT. | |
| Novelty Judge Bench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NPCBench | Professional work | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Qwen3.6 27B. | |
| NPHardEval | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for GPT-4 Turbo. | |
| NTCIR-13 MedWeb | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NTCIR-17 Radiology Report TNM Staging | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NTCIR-17 Social Media Adverse Drug Event Detection | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NTCIR-19 MedNLP-CALL Triage Classification | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NTREX-128 | Knowledge and language | 1 | 3 of 4 | Newest score The newest came 19 months ago, in April 2025, for Cohere Command A. | |
| Nuclear Decision-Making Benchmark | Safety and security | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. | |
| Numb3rs | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Numeric Grounding | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 9 months ago, in February 2026, for GLM-5. | |
| NV Chat Benchmark | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| NWPU-Captions | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NYT Connections (Prime Community) | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| NYU CTF Bench | Safety and security | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for GPT-4.1. | |
| NYU Depth V2 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OAB Exams | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OAB-Bench | Professional work | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for Gemini 3.1 Flash-Lite. | |
| OBI-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in August 2024, for mPLUG-Owl3-7B. | |
| ObjectNet | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OccuBench | Professional work | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for MiniMax-M2.7. | |
| OceanGym | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 15 months ago, in August 2025, for GLM-4.5V. | |
| OCR Arena | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| OCR-VQA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for GPT-4.1. | |
| OCT-Bench | Professional work | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Mini. | |
| OctoCodingbench | Coding | 1 | 10 of 11 | Newest score The newest came 11 months ago, in December 2025, for MiniMax-M2.1. | |
| OdysseyArena | Agents and tools | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. | |
| OdysseyArena AI Trading | Agents and tools | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. | |
| OdysseyArena Energy Dispatch | Agents and tools | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. | |
| OdysseyArena Repo System | Agents and tools | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. | |
| OdysseyArena Turn on Lights | Agents and tools | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. | |
| OdysseyBench | Professional work | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| OfficeQA Pro V2 | Professional work | 1 | 9 of 10 | Known issues 1 finding disputed or fixed only in a copy: Broken grader, leaked answers, unclear tasks and trips content filters. | |
| OGS-Bench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| OHSUMED | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OIBench | Coding | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for LongCat-Flash-Thinking-2601. | |
| OK-VQA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for Gemma 3 12B. | |
| Okapi MMLU | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 16 months ago, in July 2025, for Falcon-H1-34B. | |
| Okapi TruthfulQA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OlymMATH | Math and science | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-0.6B. | |
| OlympiadBench | Math and science | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for InternVL3.5-241B-A28B. | |
| OMEGA | Math and science | 1 | 4 of 5 | Newest score The newest came 11 months ago, in December 2025, for Olmo 3.1 32B Think. | |
| Omni-Cloze | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Omni-InteractScope | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Gemini 3.5 Flash. | |
| Omni-InteractScope Explicit Request | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Gemini 3.5 Flash. | |
| Omni-InteractScope Proactive | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Gemini 3.5 Flash. | |
| Omni-MATH | Math and science | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Ring-1T. | |
| OmniAI OCR Benchmark | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for STEP3-VL-10B. | |
| OmniBench | Vision, audio and robotics | 1 | 10 of 11 | Newest score The newest came 13 months ago, in October 2025, for LongCat-Flash-Omni. | |
| OmniCAD | Professional work | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| OmniDocBench | Vision, audio and robotics | 1 | 3 of 4 | Importer The record links a board on github.com that we don’t read yet. | |
| OmniEarth-Bench | Math and science | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o. | |
| OmniEchoBench | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| OmniGameArena | Agents and tools | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| OmniGUI | Agents and tools | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. | |
| OmniLayout | Professional work | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| Omnilingua-Bench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Omnilingual ASR Corpus | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OmniMech | Professional work | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| OmniMedVQA | Professional work | 1 | 9 of 10 | Newest score The newest came 3 years ago, in May 2023, for instructblip-vicuna-7b. | |
| OmniPhys | Math and science | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| OmniSch | Professional work | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-122B-A10B. | |
| OmniScore-Bench | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| On Evaluating User Simulators for Interactive Agents | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Online Experience (Moonshot) | Professional work | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| OOLONG | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 31B IT. | |
| Open Arabic LLM Leaderboard v2 | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Falcon-Arabic-7B. | |
| Open ASR Leaderboard | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Open Chinese LLM Leaderboard | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in June 2024, for Qwen2-0.5B. | |
| Open CoT Leaderboard | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for DeepSeek-R1-Distill-Llama-70B. | |
| Open FlagEval VLM Leaderboard | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for InternVL3-78B. | |
| Open Hebrew LLM Leaderboard | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| Open Japanese LLM Leaderboard | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for Reka Flash 3.1. | |
| Open Ko-LLM Leaderboard 2 | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in November 2024, for Qwen2.5-Coder-14B-Instruct. | |
| Open LMM Reasoning Leaderboard | Other | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for SAIL-VL-1.6-8B. | |
| Open Persian LLM Leaderboard | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for Gemma 3 12B. | |
| Open Portuguese LLM Leaderboard | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for Keye-VL-1.5-8B. | |
| Open Prompt Injection | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Open RAG Benchmark | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Open Tabletop GM Narrative Quality Probe v2 | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| Open Tabletop GM Session Command Probe | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 31B IT. | |
| Open TTS Leaderboard 1.6.0 | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 10 months ago, in January 2026, for Qwen3-TTS-12Hz-1.7B-Base. | |
| OpenAI performance in cases flagged by users | Knowledge and language | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| OpenAI user-flagged hallucination eval | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| OpenAudioBench AlpacaEval | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for Baichuan-Omni-1.5. | |
| OpenAudioBench Llama Questions | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for Baichuan-Omni-1.5. | |
| OpenAudioBench Reasoning QA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for Baichuan-Omni-1.5. | |
| OpenAudioBench TriviaQA | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for Baichuan-Omni-1.5. | |
| OpenAudioBench Web Questions | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for Baichuan-Omni-1.5. | |
| Opencpop | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OpenDesign Arena | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| OpenEarthAgent | Agents and tools | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| OpenEQA | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for gemini-robotics-er-1.5-preview. | |
| OpenFARM Bioacoustics | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OpenFinData | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OpenHands Index | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| OpenHuEval | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OpenLMM Spatial Leaderboard | Other | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| OpenLMM Subjective Leaderboard | Other | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Gemini 2.5 Flash. | |
| OpenMed MedReason Evaluation | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OpenRCA | Coding | 1 | 4 of 5 | Newest score The newest came 9 months ago, in February 2026, for Claude Sonnet 4.6. | |
| OpenRSI Index | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OpenThoughts-TBLite | Agents and tools | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for Kimi K2.5. | |
| OpenTTD-Bench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OpenVLM Leaderboard | Other | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| OpenVLM Video Leaderboard | Other | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for MiMo-VL-7B-RL. | |
| OphthoQuestions ophthalmology board preparation | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OpinionQA | Safety and security | 1 | 8 of 9 | Newest score The newest came 4 years ago, in November 2022, for davinci-003. | |
| OPT-BENCH | Coding | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for DeepSeek-V3.1. | |
| OPT-RSVG | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OptimBench | Coding | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Opusparcus | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OR-License | Professional work | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Orak | Agents and tools | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for Nemotron 3 Nano 4B. | |
| OrangeSum | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Nemotron 3 Nano 4B. | |
| ORCA-bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| OS-Critic Bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Claude Sonnet 4.5. | |
| OS-Kairos | Agents and tools | 1 | 9 of 10 | Newest score The newest came 2 years ago, in August 2024, for Qwen2-VL-7B-Instruct. | |
| OSGym | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OST-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for GPT-4.1. | |
| OSWorkerBench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| OSWorld-MCP | Agents and tools | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. | |
| OSWorld-Science | Agents and tools | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| OTelBench | Coding | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GLM-4.7. | |
| OTIS Mock AIME 2024-2025 | Math and science | 1 | 7 of 8 | Importer The record links a board on epoch.ai that we don’t read yet. | |
| OverSearchQA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Oversight gaming | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| OvertonBench | Safety and security | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for DeepSeek-R1. | |
| OVO-S-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| Oxford Flowers 102 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Oxford-IIIT Pet Dataset | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| P3B3 | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 31B IT. | |
| PACT (robot arm orchestration) | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PAgentBench | Agents and tools | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Paired-symmetry suite | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Pairwise safety red teaming | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Palmyra X6 Enterprise Workflow Evaluations | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PaperBench Code-Dev | Coding | 1 | 8 of 9 | Newest score The newest came 23 months ago, in December 2024, for o1. | |
| PaperScope | Agents and tools | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for GPT-5.1. | |
| PARADE Audio | Safety and security | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Gemini 2.5 Flash. | |
| Parafraseja | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Parametric CAD Bench V1 | Professional work | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| ParlaCAP Test Datasets | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash-Lite. | |
| ParlaSent | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| Parley Labs negotiation tasks | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ParliamentBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| PARTNR | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PASCAL VOC 2007 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PASCAL VOC 2012 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PASCAL-Context | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Past IMO (Seed-Prover formal set) | Math and science | 1 | 9 of 10 | Newest score The newest came 16 months ago, in July 2025, for Seed-Prover. | |
| PatchBench (OpenAI) | Safety and security | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| PathMCQA (Google) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Pattern Labs cyberoffensive challenges | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PatternEval | Knowledge and language | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| PazaBench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PazaBench v2 | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 21 months ago, in February 2025, for Phi-4-multimodal. | |
| PCA-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in March 2024, for DeepSeek-VL-7B-Chat. | |
| PCB-Bench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PCBWorld-Bench | Professional work | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Mini. | |
| PDEAgent-Bench | Coding | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7. | |
| PE-A-Frame internal audio event localization | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PE-AV internal video-to-audio retrieval | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Peer Arena (oddbit.ai) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| PeerQA | Math and science | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for deepseek-r1-0528-qwen3-8b. | |
| Pencil Puzzle Bench | Agents and tools | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| PenguinScrolls | Knowledge and language | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| People Making Conversation | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for Gemini 3.1 Flash Live. | |
| Per-DOC | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in July 2023, for Llama 2-13B. | |
| Per-MPST | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 3 years ago, in July 2023, for Llama 2-13B. | |
| Perception Test MCVQA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 16 months ago, in July 2025, for Gemini 2.5 Flash. | |
| PerMMLU | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| Persian NLG | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| PersistBench | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Person Identification evaluation | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Personal Agent Bench | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Personal data evaluations | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PersonaShot | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for LTX-2.3. | |
| Petri | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Petri 2.0 | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PFBench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PGG-Bench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for Gemini 2.5 Pro. | |
| PhAIL DROID: Everyday tasks | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| PHEB | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PhiBench | Other | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| PHMForge | Professional work | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Claude Opus 4.6. | |
| PhoneHarness Bench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Seed2.0 Pro. | |
| PHYRE | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PhyRubric-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 16 months ago, in July 2025, for Wan2.2-T2V-A14B. | |
| PhysGym Arena DR-Hard | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Physics GRE (Inflection) | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Physiome-ODE | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PhysVista | Vision, audio and robotics | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Pico DSL v0 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Pieverse Agent Prediction Arena | Math and science | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| Pigeonholing Partisans four-word generation | Math and science | 1 | 9 of 10 | Newest score The newest came 2 years ago, in June 2024, for Gemma 2 27B. | |
| PII Masking | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Claude Sonnet 4.5. | |
| PII-Masking-300k | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PinchBench | Agents and tools | 1 | 9 of 10 | Lab claims 1 of 3 pairs differs by more than 10 points, the largest for GLM-5V-Turbo (+14 pts). | |
| PiRhDy Melody Completion | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PISA-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-4B-Instruct. | |
| PitchBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. | |
| PLABA 2023 Abstract Adaptation | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PLABA 2024 Complete Abstract Adaptation | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PLABA 2024 Term Replacement | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Places205 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PlanBench | Math and science | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for DeepSeek-R1. | |
| PLM-VideoBench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 21 months ago, in February 2025, for Gemini 2.0 Flash. | |
| PlotPoints Round 1: Single-Turn Arena | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| PMBench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Claude Sonnet 4.6. | |
| PMPP CUDA Evaluation | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PMPP-Hard | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Poem Sentiment | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PoETa v2 | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-1.7B. | |
| Point-Bench | Vision, audio and robotics | 1 | 10 of 11 | Newest score The newest came 8 months ago, in March 2026, for Seed1.8. | |
| PokeAgent Challenge | Agents and tools | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| Poker Arena | Agents and tools | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| Polaris-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| PolEmo 2.0 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Policy Agent | Professional work | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| Polish Medical Exams | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for Llama 3.2 1B. | |
| PopCS | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PopQA | Knowledge and language | 1 | 5 of 6 | Newest score The newest came 11 months ago, in December 2025, for Olmo 3.1 32B Think. | |
| Portal2Bench | Agents and tools | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| PosteriorBench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PowerCodeBench | Coding | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. | |
| PPLX Online LLM Human Evaluation | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PPTCoEditBench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PPTFixBench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Pravda data voids chatbot audit | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Flash. | |
| PRDBench | Coding | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GLM-4.7. | |
| Prediction Arena | Math and science | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| Preference Proxy Evaluations | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for Skywork-Reward-Gemma-2-27B. | |
| Preference-ASR | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 21 months ago, in February 2025, for Phi-4-multimodal. | |
| PreScience | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PretrainGym | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PRISM Instruction Order Sensitivity | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash-Lite. | |
| PRISM-Bench (image generation) | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for BAGEL. | |
| PRMBench | Math and science | 1 | 8 of 9 | Newest score The newest came 21 months ago, in February 2025, for Gemini 2.0 Flash. | |
| ProgramBench | Coding | 1 | 11 of 12 | Lab claims 1 of 6 pairs differs by more than 10 points, the largest for Claude Opus 5 (+18 pts). | |
| ProjectTest | Coding | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| Prompt injection test suite for LLM-based machine translation | Safety and security | 1 | 3 of 4 | Newest score The newest came 14 months ago, in September 2025, for Command A Translate. | |
| Proof News election misinformation debunking | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in June 2023, for GPT-4. | |
| ProofNet | Math and science | 1 | 4 of 5 | Newest score The newest came 19 months ago, in April 2025, for DeepSeek-Prover-V2-671B. | |
| ProofRank | Math and science | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| ProOPF-B | Coding | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| PropensityBench | Safety and security | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| ProSA | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in February 2024, for Qwen1.5-14B-Chat. | |
| PROST | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 7 years ago, in November 2019, for GPT-2 (1.5B). | |
| Prosus Vending Bench v2 | Professional work | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| ProteinLMBench | Safety and security | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for Intern-S1. | |
| ProtocolLLM | Professional work | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Qwen3.6 27B. | |
| ProverBench (DeepSeek) | Math and science | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for DeepSeek-Prover-V2-671B. | |
| ProxyWar | Agents and tools | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Flash. | |
| PSEBench | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| PsychEthicsBench | Safety and security | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for Qwen2.5-14B. | |
| PsychoBench | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PsyTAR | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PTCG-Bench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. | |
| Public Tests, Hidden Tests | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PubMed-PICO | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| PubMedQA | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for Reka Core. | |
| PunEval | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in February 2024, for Claude 3 Opus. | |
| Putnam 2024 | Math and science | 1 | 2 of 3 | Confidence Likely. Some facts rest on secondary sources. | |
| Putnam 2025 | Math and science | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Putnam-AXIOM | Math and science | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for o1-preview. | |
| PuzzleWorld | Agents and tools | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| Py Bug Trace Level 3 | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Q2D-Web | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Qwen3-Embedding-0.6B. | |
| QA4MRE | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| QALB 2014 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| QALB 2015 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Qasper | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for Phi-4 Mini. | |
| QCircuitBench | Coding | 1 | 7 of 8 | Newest score The newest came 2 years ago, in April 2024, for Llama 3-8B. | |
| QIVD | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| QuAC | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in November 2024, for Llama 3.1-405B. | |
| Qualcomm Interactive Cooking | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| QuantEval | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| QuantSightBench | Math and science | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| QuixBugs | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| QVHighlights | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Nova 2.0 Pro Preview. | |
| Qwen in-house Chinese automatic evaluation | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Qwen in-house English automatic evaluation | Other | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Qwen in-house voice-chat benchmark | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Qwen-Image-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Flash Image. | |
| Qwen-Music Cover Song Generation | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Qwen3 forced alignment benchmarks | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Qwen3-ASR internal transcription benchmarks | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| QwenBlenderBench | Coding | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| QwenChineseBench | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| QwenSVG | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| QwenVision2Code | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| QwenVisualOffice | Vision, audio and robotics | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| QwenWebDev | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| QwenWorldBench | Vision, audio and robotics | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| R-Bench (real-world robustness) | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Nemotron Nano 12B v2 VL. | |
| R-ConstraintBench | Professional work | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| RaceBias (Race-Based Medicine) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RadioBench | Professional work | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| RAFT (Real-world Annotated Few-shot Tasks) | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 5 years ago, in March 2021, for GPT-Neo-2.7B. | |
| Railhead | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Ramp Stack Accounting Benchmark | Professional work | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Gemini 3.5 Flash. | |
| Range Repair Agent | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RareDiseaseBench | Math and science | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| Rauh Bundestag sentence sentiment validation | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RAVDESS | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RBench-M | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-30B-A3B-Instruct. | |
| RCM-Bench-1 | Professional work | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| RCTs to Scale nudge forecasts | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RD-TableBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Nemotron Nano 12B v2 VL. | |
| RE-Bench | Coding | 1 | 3 of 4 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. | |
| RE-Bench subset (Anthropic) | Coding | 1 | 3 of 4 | Newest score The newest came 21 months ago, in February 2025, for Claude 3.7 Sonnet. | |
| ReactSim-Bench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Read-aloud Assessment Benchmark | Vision, audio and robotics | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| REAL | Agents and tools | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Claude Opus 4. | |
| Real Point-of-Care Queries | Professional work | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| Real-scenario prompts human evaluation (Z.ai) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Real-Time Reasoning Gym | Agents and tools | 1 | 7 of 8 | Newest score The newest came 11 months ago, in December 2025, for DeepSeek-V3.2. | |
| RealBench (Verilog) | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for o1-preview. | |
| RealEstate10K | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RealHall | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| RealHarm | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RealityTest | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RealKIE-FCC Verified | Vision, audio and robotics | 1 | 4 of 5 | Newest score The newest came 11 months ago, in December 2025, for Nova 2.0 Pro Preview. | |
| REALM | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for π0-FAST. | |
| Realm Financial Reasoning | Professional work | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| RealToxicityPrompts | Safety and security | 1 | 3 of 4 | Newest score The newest came 14 months ago, in September 2025, for Apertus 70B. | |
| Reasoning behavior classifiers (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Reasoning extraction red-teaming | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| ReasonZoo | Math and science | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for DeepSeek-R1. | |
| RECODE-H | Coding | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| Recovery-Bench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Red Team Attack Simulator | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RedlineBench v2 (Fa1zvn) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RedTeam Arena | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Ref-YouTube-VOS | Vision, audio and robotics | 1 | 6 of 7 | Confidence Likely. Some facts rest on secondary sources. | |
| RefChecker | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RefuNQ | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in September 2023, for Mistral-7B-Instruct-v0.1. | |
| RefusalBench (Nous Research) | Safety and security | 1 | 10 of 11 | Newest score The newest came 11 months ago, in December 2025, for Hermes 4.3 36B. | |
| Refusals + System Jailbreak (xAI) | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Refusals + User Jailbreak (xAI) | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Regurgitation Evaluations | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ReMI | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for STEP3-VL-10B. | |
| ReplicationBench | Math and science | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Claude Sonnet 4.5. | |
| RepoInject v1 | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| RepoNormBench | Coding | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. | |
| ReportBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for Claude Sonnet 4. | |
| ReproBAIT | Safety and security | 1 | 3 of 4 | Task access We don’t know whether the tasks are public. | |
| Research Decision Benchmark | Math and science | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Research sandbagging (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Research-Eval | Agents and tools | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Grok 4.1 Fast. | |
| ResearchCodeBench | Coding | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| ResearchTasteBench | Coding | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| ReSpaceBench | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| RetailBench | Professional work | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Grok 4.3. | |
| RetryBudget-24: Policy and JSON Contracts | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Flash. | |
| REVERSAL-BENCH | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Reverse Wikipedia Paragraphs | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ReviewBench | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Reward Hack Bench | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RewardBench 2 | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Skywork-Reward-V2-Llama-3.1-8B. | |
| RExBench | Coding | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Claude Opus 4.5. | |
| RGB | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RGraph-Eval | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 6 months ago, in May 2026, for Cosmos3-Nano. | |
| RiceChem | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in April 2024, for Gemma 1.1 7B IT. | |
| RIICHI ORBIT mahjong bot comparison (open practice) | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| RIICHI ORBIT mahjong bot comparison (two-stage practice) | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| RISEBench++ | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Flash Image. | |
| RMBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for π0.5. | |
| Roark Live Bench v1.3 | Agents and tools | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| RoboARCade | Vision, audio and robotics | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| RoboArena | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for π0.5. | |
| RoboCasa-Mem | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for π0.5. | |
| RoboCasa365 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GR00T N1.6. | |
| RoboChallenge Table30 v2 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for π0.5. | |
| RoboColiseum | Vision, audio and robotics | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| RoboDx | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Roboflow 100-VL | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for Qwen2.5-VL-72B. | |
| RoboGate | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GR00T N1.6. | |
| RoboRewardBench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| RoboSpatial-Home | Vision, audio and robotics | 1 | 10 of 11 | Newest score The newest came 6 months ago, in May 2026, for Cosmos3-Edge. | |
| Robot Capabilities Index (Robocurve) | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RoboTeam | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Robotic Origami Challenge (IROS 2026) | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Robots Evolved GUI lessons | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RoboTwin 2.0 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for π0.5. | |
| robust-kbench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RoCS-MT | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RoCS-MT v2 | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for GPT-4.1. | |
| ROI Calculator (dsm) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ROK-FORTRESS | Safety and security | 1 | 3 of 4 | Task access We don’t know whether the tasks are public. | |
| ROME | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| Room-Across-Room (RxR) | Vision, audio and robotics | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| ROSTrans | Coding | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| RoTA-Bench | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| RouterBench | Other | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RPC-Bench | Math and science | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for GLM-4.7. | |
| RPEval | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Pro. | |
| RQA-70K weather radar forecast evaluation | Math and science | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for Claude 3.7 Sonnet. | |
| RSCC | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for InternVL3-8B. | |
| RSI Index (OpenAI) | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RSIBench-Data | Coding | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| RTEB | Other | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Qwen3-Embedding-4B. | |
| RTL-OPT | Professional work | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Qwen3.6 27B. | |
| RTL-Repo | Professional work | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Kimi K2. | |
| RTLLM 2.0 | Professional work | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Qwen3.6 27B. | |
| RTSGameBench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RuC-tt07-32k | Professional work | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for DeepSeek-V3.1-Terminus. | |
| RuLES 2.0 | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in February 2024, for Claude 3 Opus. | |
| RuleTrek | Knowledge and language | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Rundkast | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Runway Model Router Benchmark | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| RVMS-Bench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| S-EMBER | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| Saarland MIDI-Audio Piano Music | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SAC-Bench | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Safe response to high-severity prompts | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| SafePyramid | Safety and security | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Gemini 3.5 Flash. | |
| Safety mode controllability (Cohere) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SAHM | Professional work | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| Sakana Japanese-English translation evaluation | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SalesBench (Insurance Sales) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SalesBench-100 | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SAM 3D Body synthetic pose evaluation | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SAM 3D Objects human preference evaluation | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Samrómur 21.05 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SandboxEscapeBench | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SArena | Coding | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for Intern-S1-Pro. | |
| SAS-Bench | Professional work | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for MiMo-7B-RL. | |
| SAT (Spatial Aptitude Training) | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for gemini-robotics-er-1.5-preview. | |
| SATA-Bench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for gpt-oss-120b. | |
| SatNet | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SAUDADE-PT | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SAVE-Bench | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| SAVU-Bench | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| SC-CoMIcs | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SCADBench | Professional work | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. | |
| ScanBench | Professional work | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| ScanNet | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| scBench (LatchBio SingleCellBench) | Math and science | 1 | 11 of 12 | Confidence Likely. Some facts rest on secondary sources. | |
| ScholarQABench | Math and science | 1 | 8 of 9 | Newest score The newest came 2 years ago, in April 2024, for Llama 3-70B. | |
| Sci Pipeline v1 | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SciArena | Math and science | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Olmo 3.1 32B Instruct. | |
| SciBench | Math and science | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for Llama 3-8B. | |
| Science Sim Chemistry | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ScienceAgentBench | Math and science | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for o1-preview. | |
| ScienceBenchmark | Coding | 1 | 8 of 9 | Newest score The newest came 3 years ago, in August 2023, for T5-Large. | |
| ScienceBoard | Agents and tools | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SciFact | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SciKnowEval | Math and science | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for Claude Sonnet 4. | |
| SciMMIR | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 3 years ago, in November 2023, for mPLUG-Owl2. | |
| SciQ | Math and science | 1 | 6 of 7 | Confidence Likely. Some facts rest on secondary sources. | |
| ScopeCheck: evidence before numbers | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Nano. | |
| SCORE (Consistency and Robustness) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ScrabbleBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| Screen2Words | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ScreenASR-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Doubao Seed 2.0 Lite. | |
| ScreenSuite | Agents and tools | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for Qwen2.5-VL-32B. | |
| ScribbleEdit | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SCROLLS | Knowledge and language | 1 | 6 of 7 | Confidence Likely. Some facts rest on secondary sources. | |
| SCT-Bench | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in September 2024, for o1-preview. | |
| SCUBA | Agents and tools | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| SD-QA | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SDBench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SDR-Bench | Professional work | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Mini. | |
| SEAL Adversarial Robustness | Safety and security | 1 | 8 of 9 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-405B. | |
| SEAL Coding | Coding | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for o3-mini. | |
| SEAL Multilingual | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for o3-mini. | |
| SEAL Precise Instruction Following | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for DeepSeek-R1. | |
| SEAL Showdown | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5 Chat. | |
| Seal-0 | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-122B-A10B. | |
| SEC filing claim verification (400 items) | Knowledge and language | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| SEC-bench | Safety and security | 1 | 8 of 9 | Newest score The newest came 21 months ago, in February 2025, for Claude 3.7 Sonnet. | |
| SEC-bench Pro 260617 | Safety and security | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| SecretPatch | Safety and security | 1 | 4 of 5 | Newest score The newest came 2 years ago, in July 2024, for Gemma 2 27B. | |
| SecureCodeReview (xAI) | Safety and security | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| SecureSWE (OpenAI) | Safety and security | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Security Configuration Verification | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Sedgewick and Wayne Algorithms Exercises | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SEED-Bench | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SeedBench (Seed Science) | Math and science | 1 | 7 of 8 | Newest score The newest came 20 months ago, in March 2025, for DeepSeek-V3. | |
| Seeing What Execution Needs | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Self-harm refusals (xAI) | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Self-Instruct User-Oriented Instructions | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Self-proliferation challenges (Phuong et al. 2024) | Safety and security | 1 | 3 of 4 | Newest score The newest came 2 years ago, in July 2024, for Gemma 2 27B. | |
| SemClinBr | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SemEval 2014 Task 7 Analysis of Clinical Text | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SemEval 2015 Task 14 Analysis of Clinical Text | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SemEval 2022 Task 8 multilingual news article similarity | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for Qwen3-Embedding-0.6B. | |
| SemiKongBench-100 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SeniorTalk | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 3 years ago, in November 2023, for Whisper large-v3. | |
| Sentences Involving Compositional Knowledge | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SentinelBench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Sequential Question Answering | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SF20K-Test | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for Pixtral 12B. | |
| SGP-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for InternVL3.5-241B-A28B. | |
| ShARe/CLEF eHealth 2013 Task 1 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SheetBench | Professional work | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| Sherlock Holmes Q&A | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| Shopping MMLU | Professional work | 1 | 8 of 9 | Newest score The newest came 2 years ago, in April 2024, for Llama 3-70B. | |
| Short-horizon computational biology tasks | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SHREC 2026 Human Perceived Visual Complexity of 3D Shapes | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SHROOM-CAP | Knowledge and language | 1 | 7 of 8 | Newest score The newest came 15 months ago, in August 2025, for gpt-oss-20b. | |
| SIB-200 | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in June 2023, for GPT-3.5 Turbo. | |
| SIGHT (Search and Intent Gauge for Hospitality and Tourism) | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SignBD-Word | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Silent Exit | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SimBench (PyChrono) | Coding | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Pro. | |
| Simple Cooccurrence Bias | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SimpleQA Hallucination Rate | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for GPT-5. | |
| SimpleQA Verified | Knowledge and language | 1 | 10 of 11 | Lab claims 1 of 10 pairs differs by more than 10 points, the largest for DeepSeek-V4-Pro (+11 pts). | |
| Simreal ML Benchmark | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SimRepairBench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SimulatorArena | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| SimulCost | Math and science | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| Singapore Factuality | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Singapore POFMA Statements | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Singapore Safety Questions | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Singapore Undesirable Content | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SkateBench v2 | Knowledge and language | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| Skill Reward Hacking | Safety and security | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for MiMo-V2.5-Pro. | |
| SLDBench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Slitherlink (savi) | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Slop Score | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Claude Sonnet 4.5. | |
| SLPHelm UltraSuitePlus | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SLR-Bench | Math and science | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| Small Rewriter Benchmark | Professional work | 1 | 8 of 9 | Newest score The newest came 3 years ago, in November 2023, for GPT-4V. | |
| SmartPlay | Agents and tools | 1 | 8 of 9 | Newest score The newest came 3 years ago, in July 2023, for Llama 2-13B. | |
| SMDD-Bench | Safety and security | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for MiniMax-M2.7. | |
| SMEBench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SMM4H 2025 Shingles Vaccine Reactions | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SMM4H-HeaRD 2026 Pathology Report Staging | Professional work | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Mini. | |
| SMolInstruct | Safety and security | 1 | 4 of 5 | Newest score The newest came 8 months ago, in March 2026, for Intern-S1-Pro. | |
| SMS Spam Collection | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SnorkelFinance | Professional work | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Snowflake Analytical Search: ArXiv Papers | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Snowflake Arctic Enterprise Intelligence | Other | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Snowflake Financial Filings RAG Evaluation | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Snowflake Financial PDF Table Parsing | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Snowflake Multimodal PDF Retrieval Evaluation | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Snowflake Text-to-SQL HARD | Coding | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7. | |
| SO-Bench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Social Science Experiment Treatment Effect Archive | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SocialMaze | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SocioHack | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SoCRATES | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Soft Bias (Internal) | Safety and security | 1 | 2 of 3 | Confidence Likely. Some facts rest on secondary sources. | |
| SokoBench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Solubility Optimization (anindya) | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SolveEdit | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Flash Image. | |
| Something-Something V2 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for V-JEPA 2. | |
| Soohak | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SOP-Bench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SoSBench | Safety and security | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| SoundingEarth audio and image retrieval (MTEB) | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| South Asian Music Understanding | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for Gemma 3 1B. | |
| SPA-Bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 2 years ago, in December 2023, for CogAgent. | |
| SPACE | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SpaCE-10 | Vision, audio and robotics | 1 | 10 of 11 | Newest score The newest came 14 months ago, in September 2025, for Qwen3-VL-235B-A22B-Instruct. | |
| SpaceDG-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Qwen 3.6 35B-A3B. | |
| SpaceXAI MTS Eval | Coding | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Sparse Fourier Recovery | Coding | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Mini. | |
| Spatial Agent Index v1 | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Spatial Intelligence Index v1 | Other | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Spatial-Gym | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SpatialBench (original study) | Safety and security | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| SpatialBench (Ropedia) | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SpatialGenEval | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SpatialViz-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for STEP3-VL-10B. | |
| SpatialWorld Digital | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| SpatialWorld Physical | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| Speaker identification | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Spec Gaming Review | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Spec2Control | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SpecDecode RL | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Speech Arena | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for GPT-4o Mini TTS. | |
| Speech-to-Speech Index | Vision, audio and robotics | 1 | 3 of 4 | Task access We don’t know whether the tasks are public. | |
| SpeechMMLU | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Speko LLM Benchmark for Voice Agents v3 | Agents and tools | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Speko Speech-to-Speech Benchmark (10-turn call) | Agents and tools | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Speko Speech-to-Speech Benchmark (20-turn concierge call) | Agents and tools | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Speko Speech-to-Speech Benchmark (six concierge scenarios) | Agents and tools | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SPI (security patch identification) | Safety and security | 1 | 4 of 5 | Newest score The newest came 2 years ago, in July 2024, for Gemma 2 27B. | |
| Spider 2.0-AIFunc | Coding | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| Spiral-Bench v1.2 | Safety and security | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| SpireBench (HermesBridge trial-v0) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| SpireClimb | Agents and tools | 1 | 8 of 9 | Importer The record links a board on epoch.ai that we don’t read yet. | |
| SpireClimb StS2 A10 | Agents and tools | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| SPOT | Math and science | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for o3. | |
| SPOT5 daily photograph scheduling | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SpotSoundBench | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Spyfall study | Agents and tools | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-8B. | |
| SQAD | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SQL-API-Bench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SQuAD v2 | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in November 2024, for Llama 3.1-405B. | |
| SRE-Bench | Safety and security | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Sregym env | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ST-VQA | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| StableToolBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 2 years ago, in July 2024, for GPT-4o mini. | |
| Stack EOL Audit: geol CLI vs API | Agents and tools | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| STAGE Cross-Scene Narrative Evolution Reasoning | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| Standard jailbreaks (xAI) | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Stanford Cars | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| STAR situated video reasoning | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| STARE | Vision, audio and robotics | 1 | 4 of 5 | Newest score The newest came 10 months ago, in January 2026, for STEP3-VL-10B. | |
| STATE-Bench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for GPT-5.5. | |
| StateBench | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| StateBench (video continuation) | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 16 months ago, in July 2025, for Wan2.2-I2V-A14B. | |
| Static Jailbreak Evaluations | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| SteerBench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Step Race | Agents and tools | 1 | 8 of 9 | Newest score The newest came 11 months ago, in December 2025, for Mistral Large 3. | |
| Step-Caption | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Step-DU | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Step-SPQA | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Step-SWE-Bench | Coding | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| StepCodeBench | Coding | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| StepEval-Audio-360 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| StepUp gym attendance forecasts | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| StereoTales | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| STL-10 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| StockBench | Professional work | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Kimi K2. | |
| StreamingQA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Structured expert probing campaign: chem-bio novel design | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Structured expert probing campaign: radiological and nuclear | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Structured Output Benchmark | Knowledge and language | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Structured Web Data Extraction | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| StudyArena Audio | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Gemini 3.1 Flash TTS. | |
| StyleInstructCaps | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Sudoku-Bench | Math and science | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for LongCat-Flash-Thinking. | |
| Suicide and self-harm evaluations (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Summ-it Coreference | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Summary of a Haystack | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in March 2024, for Claude 3 Haiku. | |
| SUN397 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Sunbird African Language Technology | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SUNRGBD | Vision, audio and robotics | 1 | 4 of 5 | Newest score The newest came 8 months ago, in March 2026, for Qwen3.5-2B. | |
| SuperCLUE Arena Text to Image | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 3 years ago, in November 2023, for Claude 2.1. | |
| SuperCLUE-AdsVU | Vision, audio and robotics | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Agent | Agents and tools | 1 | 8 of 9 | Newest score The newest came 2 years ago, in June 2024, for Qwen2-72B. | |
| SuperCLUE-Agent Scenario | Agents and tools | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-AISearch | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-ASR | Vision, audio and robotics | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Code Scenario | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Coder | Coding | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-ComicShorts | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-CPIFOpen | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SuperCLUE-DeepResearch | Agents and tools | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-DeepSWE | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Edit | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Fin | Professional work | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-I2V | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Image | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-LarkClaw | Agents and tools | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Law | Professional work | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-LongContext | Knowledge and language | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-MedAssist | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Office | Professional work | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-OnDevice | Knowledge and language | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-R1 Search | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-R2V | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-RAG | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in May 2024, for GPT-4o. | |
| SuperCLUE-Safety | Safety and security | 1 | 8 of 9 | Newest score The newest came 3 years ago, in November 2023, for Yi-34B. | |
| SuperCLUE-SWE | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-SWE with Claude Code | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-SWE with mini-swe-agent | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Terminal | Agents and tools | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-TTS | Vision, audio and robotics | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-VideoEdit | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-VLM | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-Voice (audiovisual interaction) | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-WeClaw | Agents and tools | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-World | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperCLUE-XClaw | Agents and tools | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| SuperLim 1.0 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Surge AI blind coding evaluation | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SUST Bangla Emotional Speech Corpus | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SusVibes | Safety and security | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| SV-Gap Reset Release | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SveaKnow Culture | Knowledge and language | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SWE agentic fabrication eval | Knowledge and language | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| SWE-bench Multilingual | Coding | 1 | 10 of 11 | Lab claims 1 of 6 pairs differs by more than 10 points, the largest for DeepSeek-V3.2 (+11 pts). | |
| SWE-Bench Pro Verified | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SWE-Bench ProMax | Coding | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Claude Sonnet 4.6. | |
| SWE-bench Verified | Coding | 1 | 9 of 10 | Lab claims 1 of 42 pairs differs by more than 10 points, the largest for Devstral Small 2 (+12 pts). | |
| SWE-bench Verified hard subset (Meta) | Coding | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| SWE-bench-Live Verified | Coding | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Claude Sonnet 4.5. | |
| SWE-CI | Coding | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| SWE-Game | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SWE-gen JS | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SWE-Lancer | Coding | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Seed2.0 Pro. | |
| SWE-Lancer Diamond (SWE Manager) | Coding | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for o3. | |
| SWE-Marathon-Ext | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| SWE-Perf | Coding | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for MiniMax-M2.1. | |
| SWE-PolyBench | Coding | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for DeepSeek-R1. | |
| SWE-rebench V2 | Coding | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for MiniMax-M2.1. | |
| SWE-Review | Coding | 1 | 10 of 11 | Newest score The newest came 11 months ago, in December 2025, for MiniMax-M2.1. | |
| SWE-smith | Coding | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for Claude Sonnet 4. | |
| SWE-Together | Coding | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| SWebBench | Agents and tools | 1 | 4 of 5 | Confidence Likely. Some facts rest on secondary sources. | |
| Swedish Benchmarks V1.0 | Knowledge and language | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Swedish Benchmarks V1.1 | Knowledge and language | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| SweParaphrase 1.0 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SweSAT Synonyms 1.0 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Swiss Job Market Monitor ICT Term Recognition | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Swiss Smartvote 2023 party agreement | Safety and security | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Qwen 3.6 Plus. | |
| Swivuriso | Vision, audio and robotics | 1 | 7 of 8 | Newest score The newest came 2 years ago, in October 2024, for Whisper large-v3-turbo. | |
| SWM-Bench (Social World Models) | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Sycophancy prompts (Anthropic) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Sydney-Captions | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SympTEMIST | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| SYNERGY+ | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Synthetic Contextual ASR Benchmark (Indic) | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| T2Ranking | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TAAROFBENCH | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for DeepSeek-V3. | |
| TableBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| TAC 2017 Adverse Drug Reaction Extraction from Drug Labels | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TACoS | Vision, audio and robotics | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| TailOR | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Taiwanese Mandarin Language Understanding | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for GPT-4 Turbo. | |
| Talk to Act | Agents and tools | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Gemma 4 E2B. | |
| TamperBench | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TarantuBench | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TartanAir v2 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Task AW IT support | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Task-dependent output homogenization evaluation | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| TaskBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 2 years ago, in January 2024, for GPT-3.5 Turbo. | |
| TASTE | Coding | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| TAT-DQA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in May 2024, for Gemini 1.5 Pro. | |
| Tatoeba Translation Challenge | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TAU Urban Acoustic Scenes 2022 Mobile | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| tau-bench Airline | Agents and tools | 1 | 8 of 9 | Newest score The newest came 2 years ago, in October 2024, for Claude 3.5 Sonnet (October 2024). | |
| tau-bench Retail | Agents and tools | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for gpt-oss-120b. | |
| Tau2 Bench Airline With Code Agents | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Tau2 Bench Verified Airline With Code Agents | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| tau2-bench Verified | Agents and tools | 1 | 10 of 11 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| TaxoBench | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TCS-Bench | Math and science | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Te Hiku Media Māori speech recognition golden dataset | Vision, audio and robotics | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| TEACh | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TeachArena | Professional work | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| TeachObs | Professional work | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Grok 4.3. | |
| TED-LIUM | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TELBench | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| TeleTables | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. | |
| TEM-VQA | Math and science | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| Tennis & Padel State Tracking Benchmark | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| Terminal-Bench | Agents and tools | 1 | 6 of 7 | Importer The record links a board on tbench.ai that we don’t read yet. | |
| Terminal-Bench 2.0 Verified (Z.ai) | Agents and tools | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for GLM-5. | |
| Terminal-Bench 4.0 | Agents and tools | 1 | 11 of 12 | Lab claims 1 of 13 pairs differs by more than 10 points, the largest for MiMo-V2.6 Flash (+13 pts). | |
| Terminal-Bench Challenges | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Terminal-Bench Data | Agents and tools | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| Terminal-Bench Pro | Agents and tools | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5 397B-A17B. | |
| Terminal-X | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| TerminalWorld | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| TerraFirm REIT | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TerretaQA | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Testing Relational Understanding in Text-Guided Image Generation | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Text Classification (ByteDance Seed in-house) | Professional work | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| Text persuasion study on political opinions | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Text to Text Safety | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Text2CAD | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TextCaps | Vision, audio and robotics | 1 | 3 of 4 | Newest score The newest came 6 months ago, in May 2026, for Gemini Embedding 2. | |
| TextQuests | Agents and tools | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| TextWorld Cooking (GATA) | Agents and tools | 1 | 7 of 8 | Newest score The newest came 15 months ago, in August 2025, for Qwen3-4B-Thinking-2507. | |
| TGVE+ | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ThaiExam | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in August 2024, for Command R+. | |
| The Aggregate Asteroid Numbers | Knowledge and language | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| The Aggregate Day-one Recall Check | Knowledge and language | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| The Aggregate Exotic Fact Recall | Knowledge and language | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| The Aggregate Forecasting Resolved Questions | Math and science | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| The Aggregate Inversion and Execution Ladder | Agents and tools | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| The Aggregate Long-tail Book Recall | Knowledge and language | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| The Aggregate Puzzle Portfolio Ladder | Agents and tools | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| The Aggregate Trip Outcome Prediction | Knowledge and language | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| The AI Teacher Test | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| The Amazing Agent Race | Agents and tools | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Mini. | |
| The GenderQueer Test Suite | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| The Illusion of AI Safety | Safety and security | 1 | 7 of 8 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| The Impossible Refund | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| The Silicon Sample Benchmark | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TheAgentCompany | Professional work | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for MiniMax-M2. | |
| TheoremQA | Math and science | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for Qwen2.5-72B. | |
| Thinkingbox-bench | Agents and tools | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Third PASCAL Recognizing Textual Entailment Challenge | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Time Horizon Index: KSP | Agents and tools | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| tinyBenchmarks | Other | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TIR-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Qwen 3.6 Plus. | |
| TissueMicroscopyBench | Math and science | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| TituLLMs Bangla Evaluation Suite | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TOEFL11 Essay Proficiency Scoring | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TokSuite | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Tone | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| ToolBench | Agents and tools | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| ToolcallFormatIFBench | Agents and tools | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| ToolComp | Agents and tools | 1 | 9 of 10 | Newest score The newest came 20 months ago, in March 2025, for Gemini 2.5 Pro. | |
| ToolMaze | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| ToolQA | Agents and tools | 1 | 9 of 10 | Newest score The newest came 3 years ago, in July 2023, for Llama 2-13B. | |
| ToRR | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for DeepSeek-V3. | |
| ToxiGen | Safety and security | 1 | 3 of 4 | Newest score The newest came 11 months ago, in December 2025, for Olmo 3.1 32B Think. | |
| TRACE | Coding | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GLM-4.7. | |
| TraceTrial-Bench | Agents and tools | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Trading Bot Bench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Trae Agent Bench | Coding | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| TRAIL | Agents and tools | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for Gemini 2.5 Pro. | |
| TRAJECT-Bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| Trajectory Labs cyber red teaming | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Transformer Puzzles (RASPy) | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Transluce Mental Health Evaluation | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Transluce User Awareness Evaluation | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TREC 2023 Clinical Trials Track | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TREC BioGen 2024 | Professional work | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Claude Sonnet 4.6. | |
| TREC Product Search 2023 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TREC-COVID | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Trialstreamer clinical evidence extraction | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Tricky Edits of Collected and Curated Images | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Flash Image. | |
| TriPhysBench | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| TriScore | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TritonBench | Coding | 1 | 9 of 10 | Newest score The newest came 22 months ago, in January 2025, for DeepSeek-R1. | |
| Trivia-et | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TRLawBench | Professional work | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| TRUE (Text Recognition and Understanding) | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for Gemini 2.5 Pro. | |
| TrustBoundary Bench | Safety and security | 1 | 7 of 8 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| TruthfulQA-Multi | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in July 2024, for Llama 3.1-70B. | |
| TruthInsightBench | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TUA-Bench | Agents and tools | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Tulu 3 Safety Eval | Safety and security | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| Turing Agentic Environments | Professional work | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7. | |
| Turing TTS | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| TurkishMMLU | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 23 months ago, in December 2024, for Qwen2.5-14B. | |
| Turku NER Corpus | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Turku Paraphrase Corpus | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TurnBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Gemini 3.1 Flash Live. | |
| TUT Urban Acoustic Scenes 2018 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TutorCaseBench | Professional work | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| TVRBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Qwen3.6 27B. | |
| Twitter Financial News Sentiment | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Two-Hop Latent Reasoning | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| TxBench-Antibody-Development | Math and science | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| TydiQA | Knowledge and language | 1 | 4 of 5 | Importer The record links a board on ai.google.com that we don’t read yet. | |
| U18 safety evaluations | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| UAQFact | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in January 2024, for GLM-4. | |
| UAV Operator | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UAV Robust Control Hub | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UCI Adult Dataset (Project Moonshot) | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UCM-Captions | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UD Ancient Greek PTNK | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UD Ancient Hebrew PTNK | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UD Latin LLCT | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UD Latin UDante | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Uganda Cultural and Cognitive Benchmark | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UGMathBench | Math and science | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Ring-1T. | |
| UHGEval | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 2 years ago, in June 2024, for GLM-4-9B-Chat. | |
| UI-Bench | Coding | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| UI-Vision | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UI-Vision Element Grounding | Agents and tools | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for Claude 3.7 Sonnet. | |
| UI-Vision Layout Grounding | Agents and tools | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for Claude 3.7 Sonnet. | |
| UIT-ViHSD | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UIT-VSFC | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UK AISI File Download monitor evasion | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| UK Dermatology SCE Sample Questions | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ULSA synthesis action retrieval | Math and science | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UMI-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for π0.5. | |
| Uncheatable Eval | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| Ungrounded Inference evaluation | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Unhealthy Comments Corpus | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UniCode | Coding | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash. | |
| Unidentified Video Objects | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UniGenBench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Cosmos3-Nano. | |
| UniGUI-Bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for GLM-5V-Turbo. | |
| Uniqueness of Optimum | Math and science | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for gpt-oss-120b. | |
| Unit Commitment Lite | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Universal Dependencies Sanskrit Vedic | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UniVerseBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Qwen3.5-Omni-Plus. | |
| Unjustified refusals | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| UNO-Bench | Vision, audio and robotics | 1 | 10 of 11 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| Unsatz | Math and science | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| UNT Computer Science Short Answer Dataset 2.0 | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UQ | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UrbanSED | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| URGENT 2024 | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| URO-Bench | Vision, audio and robotics | 1 | 4 of 5 | Newest score The newest came 7 months ago, in April 2026, for Qwen3.5-Omni-Plus. | |
| US-DermMCQA (Google) | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| USACO | Coding | 1 | 9 of 10 | Newest score The newest came 2 years ago, in February 2024, for Claude 3 Sonnet. | |
| UserBench | Agents and tools | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. | |
| USIM underwater robot tasks | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for GR00T N1.5. | |
| USMLE Step 1 (sample questions) | Professional work | 1 | 4 of 5 | Newest score The newest came 3 years ago, in July 2023, for Claude 2. | |
| USMLE Step 2 (sample questions) | Professional work | 1 | 4 of 5 | Newest score The newest came 3 years ago, in July 2023, for Claude 2. | |
| USMLE Step 3 (sample questions) | Professional work | 1 | 4 of 5 | Newest score The newest came 3 years ago, in July 2023, for Claude 2. | |
| UW Clinical Record Section Chunker i2b2 Discharge Summary Corpus | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| UWBench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Uyghur Language Question Answering | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| V-JEPA 2 zero-shot robot manipulation | Vision, audio and robotics | 1 | 6 of 7 | Task access We don’t know whether the tasks are public. | |
| VABAR | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VAERS influenza vaccine adverse event extraction | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VagueBench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 19 months ago, in April 2025, for InternVL3-78B. | |
| VAHBench | Vision, audio and robotics | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Legal AI Report | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Vals Smith: apache/spark | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: arandomhooman/hoomans-morphe-patches | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: arduino/arduino-ide | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: astral-sh/uv | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: AumeshMisra/tensorzero | Coding | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: benredmond/apex | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: BerriAI/litellm | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: chrisanderton/tyk-gateway | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: conjfrnk/blackbird | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: conjfrnk/memovox | Coding | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: dwojtaszek/zipper | Coding | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: Effect-TS/effect | Coding | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: Expertium/fsrs-autoresearch | Coding | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: flutter/flutter | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: fmtlib/fmt | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: geopandas/geopandas | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: goharbor/harbor | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: jax-ml/jax | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: ladydascalie/core | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: leanprover-community/mathlib4 | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: Lemonawa/music163bot-rust | Coding | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: mirumee/ariadne | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: nationalsecurityagency/ghidra | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: NVIDIA/nccl | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: OpenTTD/OpenTTD | Coding | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: oven-sh/bun | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: palantir/blueprint | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: pallets-eco/flask-wtf | Coding | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: poalrom/sharepaste | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: punt-labs/lux | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: sashabaranov/go-openai | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: schubydoo/clauster | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: sgl-project/sglang | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: skypilot-org/skypilot | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: suitedaces/dorabot | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: tensorflow/tensorflow | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: tock/tock | Coding | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: torvalds/linux | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: vastsa/PI-Desktop | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: vercel/next.js | Coding | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Vals Smith: vllm-project/llm-compressor | Coding | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| Valyrian Games Coding Challenge | Coding | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5 mini. | |
| VAmoS Pro Bench | Agents and tools | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| VATEX | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Gemini Embedding 2. | |
| VATEX ZH | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 16 months ago, in July 2025, for Gemini 2.5 Flash. | |
| VBench-2.0 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Veo 3. | |
| VBVR-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 16 months ago, in July 2025, for Wan2.2-I2V-A14B. | |
| VBVR-Pro | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 16 months ago, in July 2025, for Wan2.2-I2V-A14B. | |
| VCB-Bench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VCG+ Diverse | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 3 years ago, in November 2023, for GPT-4V. | |
| VCR (Visual Caption Restoration) | Vision, audio and robotics | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| VCTK Corpus | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| vCX Benchmark | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| VDC Benchmark | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in August 2024, for LLaVA-OneVision 7B. | |
| VDEBench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Qwen-Image-Edit-2509. | |
| VDR-Bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-8B-Instruct. | |
| Vectara Hallucination Leaderboard (HHEM) | Knowledge and language | 1 | 8 of 9 | Confidence Likely. Some facts rest on secondary sources. | |
| VELOCITI | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 2 years ago, in August 2024, for Qwen2-VL-7B-Instruct. | |
| VenusBench-Mobile | Agents and tools | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for Kimi K2.5. | |
| Verbalized grader awareness (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Verbalized metagaming | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| Verifiability Judgments for Generative Search | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VerifierBench | Other | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Qwen3-14B. | |
| VerilogEval | Professional work | 1 | 9 of 10 | Newest score The newest came 2 years ago, in September 2024, for o1-preview. | |
| VeRO-Bench | Coding | 1 | 7 of 8 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| VGGSound | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VHELM | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Pro. | |
| Vibe Code Bench | Coding | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| Vibe-Eval | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Seed2.0 Pro. | |
| Video Browser Showdown 2026 marine video search | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Video SimpleQA | Knowledge and language | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Video to Valence | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Video-MonetBench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Video-Oasis | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Qwen3-VL-8B-Instruct. | |
| Video-SafetyBench | Safety and security | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for Qwen2.5-VL-32B. | |
| Video2Robot Embodied | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VideoACE | Vision, audio and robotics | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| VideoCon | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VideoEval-Pro | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Seed2.0 Lite (0428). | |
| VideoGameBench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Llama 4 Maverick. | |
| VideoGameBench Lite | Agents and tools | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for Llama 4 Maverick. | |
| VideoHallucer | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for Reka Edge. | |
| VideoPhy-2 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Cosmos3-Nano. | |
| VideoScore-Bench v2 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| VIF-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Qwen-Image-Edit-2509. | |
| ViMU | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for MiMo-V2-Omni. | |
| ViNV-TTS | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VIPBench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VIPSeg | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VisEditBench (TikZ diagram editing) | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 31B IT. | |
| Vision Arena | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| VisionArena-Bench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in October 2024, for Gemini 1.5 Flash 8B. | |
| ViSpeak-Bench | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| VIST | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VISTA | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4. | |
| VISTA (Visually Impaired Scene and Task Annotations) | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 8 months ago, in March 2026, for Gemini 3.1 Flash-Lite. | |
| ViStoryBench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 14 months ago, in September 2025, for Qwen-Image-Edit-2509. | |
| Vistra | Knowledge and language | 1 | 3 of 4 | Newest score The newest came 10 months ago, in January 2026, for TranslateGemma 27B. | |
| Visual Aesthetic Benchmark | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. | |
| Visual Strategic Bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for o3. | |
| VisualPuzzles | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for ERNIE 5.0. | |
| VisualToolBench | Agents and tools | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| VisualWebArena | Agents and tools | 1 | 9 of 10 | Newest score The newest came 2 years ago, in April 2024, for Llama 3-70B. | |
| ViVerBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Seed2.0 Pro. | |
| VizWiz | Vision, audio and robotics | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| VLA-Arena | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for MolmoAct 2. | |
| VLABench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in June 2024, for OpenVLA. | |
| VLAIR Legal Research | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VLegal-Bench | Professional work | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Claude Sonnet 4.5. | |
| VLSP 2023 Vietnamese Mispronunciation Detection | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VM14K | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Vmax Tasks | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VocalGrad | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for Gemini 3 Flash. | |
| Voice Arena Diarization Bench | Vision, audio and robotics | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| Voice Arena Dictation Bench | Vision, audio and robotics | 1 | 9 of 10 | Task access We don’t know whether the tasks are public. | |
| Voice Arena Goal Dataset | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for Qwen2.5-Omni-7B. | |
| Voice Jailbreak Attacks Against GPT-4o | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VOX-DUB | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VoxAnon | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VoxCeleb2 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Gemini 2.5 Flash. | |
| VoxelBench | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| VPCT | Vision, audio and robotics | 1 | 5 of 6 | Importer The record links a board on epoch.ai that we don’t read yet. | |
| VRBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Keye-VL-8B-Preview. | |
| VRSBench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 3 years ago, in November 2023, for GPT-4V. | |
| VSEC | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VSI-Bench-Debiased | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 19 months ago, in April 2025, for InternVL3-9B. | |
| VSR | Vision, audio and robotics | 1 | 6 of 7 | Confidence Likely. Some facts rest on secondary sources. | |
| Vulnerability finding evaluations (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| VulnGym | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| VulnLMP | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| VWE-Bench | Professional work | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| Wahl-O-Mat 2024 European election party agreement | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Wahl-O-Mat 2025 Bundestag party agreement | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WARC-Bench | Agents and tools | 1 | 9 of 10 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| Washington Post political lean evaluation | Safety and security | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Claude Opus 4.8. | |
| WASSCE Benchmark for Open Language Models | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WAXAL Speech Recognition Test | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Wayfinder Code Search Benchmark | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WBench | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 6 months ago, in May 2026, for Cosmos3-Nano. | |
| WearableQA | Professional work | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Gemma 4 26B A4B. | |
| WeatherQA (severe weather) | Math and science | 1 | 9 of 10 | Newest score The newest came 11 months ago, in December 2025, for GPT-5.2. | |
| Web Context (Nimble) | Agents and tools | 1 | 10 of 11 | Confidence Likely. Some facts rest on secondary sources. | |
| Web of lies | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WebExploitBench | Safety and security | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| WebGuard | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WebTailBench | Agents and tools | 1 | 10 of 11 | Lab claims 1 of 2 pairs differs by more than 10 points, the largest for Fara-7B (−10 pts). | |
| WebVR | Coding | 1 | 8 of 9 | Newest score The newest came 9 months ago, in February 2026, for Gemini 3.1 Pro. | |
| WeChat Pay Physical Client Evaluation | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WeGenEditBench | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| WeirdChat | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WeirdML | Coding | 1 | 8 of 9 | Task access We don’t know whether the tasks are public. | |
| WeirdML v3 | Coding | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| Werewolf Arena | Agents and tools | 1 | 8 of 9 | Newest score The newest came 2 years ago, in December 2023, for Gemini 1.0 Pro. | |
| WHAMR! | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Where2Place | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 6 months ago, in May 2026, for Cosmos3-Edge. | |
| White-box analyses of model internals (Anthropic) | Safety and security | 1 | 4 of 5 | Independent runs No scores from any source yet. | |
| WHU-CDC | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WiC | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in May 2023, for PaLM 2-L. | |
| Wiki Live Challenge | Agents and tools | 1 | 8 of 9 | Newest score The newest came 15 months ago, in August 2025, for GPT-5. | |
| Wiki Neutrality Corpus | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WIKIBIAS | Safety and security | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WikiFact (HELM) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WikiNEuRal | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Wikispeedia | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WikiVQABench | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WildArtifactBench | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WildAWM validation | Agents and tools | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Qwen3.5-4B. | |
| WildBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 10 months ago, in January 2026, for STEP3-VL-10B. | |
| WildGuardTest | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| WildKhmerST | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WildScore | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 8 months ago, in March 2026, for GPT-5.4 Mini. | |
| WildVision Arena | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 2 years ago, in April 2024, for Idefics2-8B. | |
| Windows Agent Arena | Agents and tools | 1 | 9 of 10 | Newest score The newest came 12 months ago, in November 2025, for Qwen3-VL-235B-A22B. | |
| Windsurf Arena Mode Leaderboard | Coding | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for Kimi K2.6. | |
| Winogender | Safety and security | 1 | 4 of 5 | Newest score The newest came 2 years ago, in July 2024, for Gemma 2 27B. | |
| WISE Verified | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 22 months ago, in January 2025, for Janus-Pro-7B. | |
| WISER-FindAll | Agents and tools | 1 | 9 of 10 | Newest score The newest came 21 months ago, in February 2025, for OpenAI deep research. | |
| WixQA | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WMT 2014 News Translation | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WMT 2016 News Translation | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WMT24 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WMT24 Icelandic Idiomatic Expressions | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 2 years ago, in August 2024, for Command R+. | |
| WMT24 Icelandic Proper Names | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WMT25 EEA Terminology | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 17 months ago, in June 2025, for Gemini 2.5 Pro. | |
| WMT26 Multilingual Instruction Shared Task | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WOLF | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Word-Level BISINDO | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Word-Task (Tencent) | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| WorkArena++ | Agents and tools | 1 | 8 of 9 | Newest score The newest came 2 years ago, in May 2024, for GPT-4o. | |
| WorkIndex | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| Workspace-Bench-Lite | Professional work | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Grok 4.3. | |
| WorkSurface-Bench | Professional work | 1 | 8 of 9 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| World-Class Bio | Safety and security | 1 | 10 of 11 | Task access We don’t know whether the tasks are public. | |
| WorldArena 2.0 | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 14 months ago, in September 2025, for Code World Model (CWM). | |
| WorldAuditBench | Vision, audio and robotics | 1 | 9 of 10 | Confidence Likely. Some facts rest on secondary sources. | |
| WorldCoder-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. | |
| WorldPrediction | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WorldSense | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for Seed2.0 Lite (0428). | |
| WorldSimBench | Vision, audio and robotics | 1 | 7 of 8 | Task access We don’t know whether the tasks are public. | |
| WorldState-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 16 months ago, in July 2025, for Wan2.2-I2V-A14B. | |
| Writing Style Matching (apaz) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WSI-Path | Professional work | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WSYue-eval | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| WTP-Bench | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 13 months ago, in October 2025, for Claude Haiku 4.5. | |
| X Browse (xAI) | Agents and tools | 1 | 10 of 11 | Newest score The newest came 12 months ago, in November 2025, for Grok 4.1 Fast. | |
| X-Stream | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 12 months ago, in November 2025, for Gemini 3 Pro. | |
| xAI Factuality (Hallucination) eval | Knowledge and language | 1 | 7 of 8 | Confidence Likely. Some facts rest on secondary sources. | |
| xAI sycophancy eval | Safety and security | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| xbench-DeepSearch-2510 | Agents and tools | 1 | 9 of 10 | Newest score The newest came 9 months ago, in February 2026, for Step 3.5 Flash. | |
| xbench-Marketing | Professional work | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for Gemini 2.5 Flash. | |
| xbench-Recruitment | Professional work | 1 | 8 of 9 | Newest score The newest came 18 months ago, in May 2025, for Gemini 2.5 Flash. | |
| XBOW Validation Benchmarks | Safety and security | 1 | 2 of 3 | Confidence Likely. Some facts rest on secondary sources. | |
| xChatBench | Vision, audio and robotics | 1 | 3 of 4 | Newest score The newest came 18 months ago, in May 2025, for Aya Vision 32B. | |
| xCodeEval | Coding | 1 | 8 of 9 | Newest score The newest came 2 years ago, in February 2024, for StarCoder 2 3B. | |
| XL-Sum (English) | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| xMMMU | Vision, audio and robotics | 1 | 9 of 10 | Newest score The newest came 18 months ago, in May 2025, for Aya Vision 32B. | |
| XNLI | Knowledge and language | 1 | 3 of 4 | Confidence Likely. Some facts rest on secondary sources. | |
| XQuAD | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 20 months ago, in March 2025, for Gemma 3 12B. | |
| XSTest | Safety and security | 1 | 4 of 5 | Newest score The newest came 8 months ago, in March 2026, for Seed1.8. | |
| XTD-200 | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| XYBench | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Pro. | |
| YC-Bench | Professional work | 1 | 10 of 11 | Lab claims 1 of 1 pair differs by more than 10 points, the largest for MiniMax-M3 (+$1,100,494). | |
| Yet Another Kernel Kernel | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| YoCausal | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 16 months ago, in July 2025, for Wan2.2-T2V-A14B. | |
| YouCook2 | Vision, audio and robotics | 1 | 8 of 9 | Newest score The newest came 16 months ago, in July 2025, for Gemini 2.5 Flash. | |
| YouCook2 retrieval | Vision, audio and robotics | 1 | 3 of 4 | Newest score The newest came 6 months ago, in May 2026, for Gemini Embedding 2. | |
| Zalo AI Challenge 2022 End-to-End Question Answering | Knowledge and language | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ZendoWorld | Agents and tools | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ZenML Bench | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ZeroCLUE | Knowledge and language | 1 | 8 of 9 | Newest score The newest came 3 years ago, in March 2023, for ChatGLM-6B. | |
| Zerolang Editing | Coding | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| ZIQI-Eval | Knowledge and language | 1 | 9 of 10 | Newest score The newest came 3 years ago, in September 2023, for Qwen-7B. | |
| Zoomed-in BVCC | Vision, audio and robotics | 1 | 3 of 4 | Independent runs No scores from any source yet. | |
| τ-Rec | Agents and tools | 1 | 9 of 10 | Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. | |
| (S)elected Moderation campaign prompts | Safety and security | 7 of 7 | Nothing to fix | ||
| 100Q-Hard (Anthropic) | Knowledge and language | 5 of 5 | Nothing to fix | ||
| 1H-VideoQA | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| 1X World Model Challenge | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| 2013 Native Language Identification Shared Task | Professional work | 8 of 8 | Nothing to fix | ||
| 2022 n2c2/UW Shared Task on Extracting Social Determinants of Health | Professional work | 8 of 8 | Nothing to fix | ||
| 2024 NIST GenAI Text-to-Text Discriminators | Other | 8 of 8 | Nothing to fix | ||
| 24/7 ClawBench 2.0 | Professional work | 5 of 5 | Nothing to fix | ||
| 3D Arena | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| 3D GameDev Arena | Coding | 9 of 9 | Nothing to fix | ||
| 3D SubwayBench | Coding | 9 of 9 | Nothing to fix | ||
| 3D Worlds Arena | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| 3DCodeArena | Professional work | 9 of 9 | Nothing to fix | ||
| 3DCodeBench | Professional work | 5 of 5 | Nothing to fix | ||
| 3DGen-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| 3DSceneBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| 3LM | Knowledge and language | 8 of 8 | Nothing to fix | ||
| A-OKVQA | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| A.S.E (AICGSecEval) | Safety and security | 8 of 8 | Nothing to fix | ||
| AA-3D | Coding | 4 of 4 | Nothing to fix | ||
| AA-AnalystAgent | Professional work | 5 of 5 | Nothing to fix | ||
| AA-Briefcase | Professional work | 5 of 5 | Nothing to fix | ||
| AA-Briefcase v1.1 | Professional work | 6 of 6 | Nothing to fix | ||
| AA-LCR 1.1 | Knowledge and language | 5 of 5 | Nothing to fix | ||
| AA-LCR Extended | Knowledge and language | 11 of 11 | Nothing to fix | ||
| AA-Omniscience | Knowledge and language | 5 of 5 | Nothing to fix | ||
| AA-Omniscience v1.1 | Knowledge and language | 4 of 4 | Nothing to fix | ||
| AAArena | Agents and tools | 9 of 9 | Nothing to fix | ||
| AAV capsid packaging prediction | Safety and security | 4 of 4 | Nothing to fix | ||
| AAV Capsid Packaging Prediction (OpenAI) | Safety and security | 6 of 6 | Nothing to fix | ||
| ABACUS | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| AbstentionBench | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Accelerate Bench | Coding | 11 of 11 | Nothing to fix | ||
| Access to Justice Benchmark | Professional work | 11 of 11 | Nothing to fix | ||
| Accounting Autocoding | Professional work | 8 of 8 | Nothing to fix | ||
| AceBench | Agents and tools | 9 of 9 | Nothing to fix | ||
| Acme Agent Evals | Agents and tools | 8 of 8 | Nothing to fix | ||
| Active data exfiltration, visual browser | Safety and security | 4 of 4 | Nothing to fix | ||
| Active-SWE | Coding | 10 of 10 | Nothing to fix | ||
| ActiveVision | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| ActivityNet Captions | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| AdaExploreBench | Agents and tools | 10 of 10 | Nothing to fix | ||
| Addis Scribe Benchmark | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| ADE Corpus | Professional work | 8 of 8 | Nothing to fix | ||
| ADE20K | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Adherence to its constitution (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| ADQA | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ADVANCE audio and image retrieval (MTEB) | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| AdvancedIF | Knowledge and language | 9 of 9 | Nothing to fix | ||
| AdvancedMathBench | Math and science | 5 of 5 | Nothing to fix | ||
| AEC-Bench (Nomic) | Professional work | 10 of 10 | Nothing to fix | ||
| AEC-Geometric-Bench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| AECV-Bench (2026 release) | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Aegis 2.0 | Safety and security | 3 of 3 | Nothing to fix | ||
| AfriBED | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| AfriQA | Knowledge and language | 7 of 7 | Nothing to fix | ||
| AfriSenti | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Agent harness misuse evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| Agent Initiative Benchmark | Professional work | 10 of 10 | Nothing to fix | ||
| Agent Island | Agents and tools | 9 of 9 | Nothing to fix | ||
| Agent JSK | Safety and security | 5 of 5 | Nothing to fix | ||
| Agent Memory Leaderboard | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Agent Replay Auto-Review | Safety and security | 4 of 4 | Nothing to fix | ||
| Agent Startup Bench | Professional work | 11 of 11 | Nothing to fix | ||
| Agent-Diff | Agents and tools | 8 of 8 | Nothing to fix | ||
| Agent-Only Fantasy Football League | Agents and tools | 9 of 9 | Nothing to fix | ||
| AgentBench | Agents and tools | 8 of 8 | Nothing to fix | ||
| AgentCallBench: Tool-Calling Reliability | Agents and tools | 9 of 9 | Nothing to fix | ||
| AgentDojo | Safety and security | 9 of 9 | Nothing to fix | ||
| AgentHarm | Safety and security | 9 of 9 | Nothing to fix | ||
| AgentHarm Verified (Meta) | Safety and security | 9 of 9 | Nothing to fix | ||
| Agentic BAIM-LLM Evaluation (ABLE) | Safety and security | 10 of 10 | Nothing to fix | ||
| Agentic coding honeypot monitor evasion | Safety and security | 4 of 4 | Nothing to fix | ||
| Agentic coding malicious requests (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| Agentic Credit Analysis | Professional work | 11 of 11 | Nothing to fix | ||
| Agentic harmful task refusals | Safety and security | 3 of 3 | Nothing to fix | ||
| Agentic task gaming and overreach evals | Safety and security | 4 of 4 | Nothing to fix | ||
| Agentic tasks (autonomous replication and adaptation) | Coding | 4 of 4 | Nothing to fix | ||
| Agentic Vision Bench (Moonshot) | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| AgenticFinBench | Professional work | 10 of 10 | Nothing to fix | ||
| Agentick | Agents and tools | 9 of 9 | Nothing to fix | ||
| AgenticRL-Bench | Coding | 11 of 11 | Nothing to fix | ||
| AgentIF | Knowledge and language | 10 of 10 | Nothing to fix | ||
| AgentIF (2026-10 setting) | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Agents' Last Exam | Professional work | 12 of 12 | Nothing to fix | ||
| Agents' Last Exam v1.1 | Professional work | 12 of 12 | Nothing to fix | ||
| AgentSettlementBench | Professional work | 8 of 8 | Nothing to fix | ||
| AgentSWE | Coding | 9 of 9 | Nothing to fix | ||
| AgentWorldBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| AGI-Eval Community Instruction-Tuned Language Models | Knowledge and language | 9 of 9 | Nothing to fix | ||
| AGI-Pilled Bench | Math and science | 10 of 10 | Nothing to fix | ||
| AGIEval | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Agri STT Benchmarking Dataset | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| AI AgriBench Post-Cutoff | Professional work | 8 of 8 | Nothing to fix | ||
| AI Battle Arena | Agents and tools | 9 of 9 | Nothing to fix | ||
| AI Code Review Evaluation (Augment Code) | Coding | 8 of 8 | Nothing to fix | ||
| AI Detection Benchmark | Knowledge and language | 9 of 9 | Nothing to fix | ||
| AI Engineering Lab model creation | Coding | 8 of 8 | Nothing to fix | ||
| AI FPL | Agents and tools | 8 of 8 | Nothing to fix | ||
| AI Professions Prompt Benchmark | Professional work | 9 of 9 | Nothing to fix | ||
| AI Security Leaderboard | Safety and security | 9 of 9 | Nothing to fix | ||
| AI Spreadsheet Benchmark | Professional work | 8 of 8 | Nothing to fix | ||
| AI2D | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Aider Polyglot | Coding | 9 of 9 | Nothing to fix | ||
| Aider Refactoring Benchmark | Coding | 8 of 8 | Nothing to fix | ||
| Aider-Edit | Coding | 10 of 10 | Nothing to fix | ||
| AILA Statute Retrieval | Professional work | 9 of 9 | Nothing to fix | ||
| AILuminate Safety 1.2 | Safety and security | 9 of 9 | Nothing to fix | ||
| AIME 1983-2024 | Math and science | 3 of 3 | Nothing to fix | ||
| AIME 2025 | Math and science | 9 of 9 | Nothing to fix | ||
| AIME 2026 | Math and science | 10 of 10 | Nothing to fix | ||
| AInstein-SWE-Bench | Coding | 4 of 4 | Nothing to fix | ||
| AIR-Bench 2024 | Safety and security | 9 of 9 | Nothing to fix | ||
| AIR-Bench 24.05 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| AIR-Bench Foundation | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| AIRS-Bench | Coding | 9 of 9 | Nothing to fix | ||
| AISA-ArabicFC 2026 | Agents and tools | 8 of 8 | Nothing to fix | ||
| AISHELL-1 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| AISHELL-2 | Vision, audio and robotics | 3 of 3 | Nothing to fix | ||
| AISHELL-4 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ALE-Bench | Coding | 10 of 10 | Nothing to fix | ||
| AlexandriaX 2026 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| ALFFA Swahili Broadcast News Test | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ALFRED | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Alignment faking evaluation | Safety and security | 9 of 9 | Nothing to fix | ||
| AliMeeting | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| AlpacaEval | Knowledge and language | 8 of 8 | Nothing to fix | ||
| AlpacaEval 2.0 | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Althingi Parliamentary Speech | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Alyah | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Amazon Counterfactual Detection | Knowledge and language | 9 of 9 | Nothing to fix | ||
| AMC 12 2023 | Math and science | 3 of 3 | Nothing to fix | ||
| AMCAE 2024 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| AmericasNLP 2025 Shared Task on Machine Translation Metrics for Indigenous Languages | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Amharic Speech Emotion Dataset | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| AML Bench (Veris AI) | Professional work | 11 of 11 | Nothing to fix | ||
| AMLBench v0.2 (Superjective) | Professional work | 8 of 8 | Nothing to fix | ||
| AMLworld-Compact | Professional work | 9 of 9 | Nothing to fix | ||
| AMO-Bench | Math and science | 10 of 10 | Nothing to fix | ||
| AmsterTime | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| AnalogCoder-Pro | Professional work | 8 of 8 | Nothing to fix | ||
| AnalogPortBench | Professional work | 11 of 11 | Nothing to fix | ||
| Android Agent Arena (201 tasks) | Agents and tools | 8 of 8 | Nothing to fix | ||
| Android Bench 2.0 | Coding | 11 of 11 | Nothing to fix | ||
| AndroidBench | Coding | 9 of 9 | Nothing to fix | ||
| AndroidControl | Agents and tools | 7 of 7 | Nothing to fix | ||
| AndroidWorld | Agents and tools | 9 of 9 | Nothing to fix | ||
| ANES presidential vote choice | Math and science | 7 of 7 | Nothing to fix | ||
| Angry Tweets | Knowledge and language | 11 of 11 | Nothing to fix | ||
| ANIMA | Safety and security | 9 of 9 | Nothing to fix | ||
| Animal Kingdom | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Animal-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Animals with Attributes 2 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ANLI | Knowledge and language | 5 of 5 | Nothing to fix | ||
| ANLI R1 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| ANLI R2 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| ANLI R3 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Answer Factuality (Perplexity) | Knowledge and language | 9 of 9 | Nothing to fix | ||
| AnthroBench | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Anthropic ECI (AECI) | Other | 5 of 5 | Nothing to fix | ||
| Anthropic political even-handedness eval | Safety and security | 10 of 10 | Nothing to fix | ||
| Anthropic Red Team | Safety and security | 8 of 8 | Nothing to fix | ||
| Anthropic undermining liberal democracy (behavioral audit metric) | Safety and security | 5 of 5 | Nothing to fix | ||
| Antidote | Knowledge and language | 9 of 9 | Nothing to fix | ||
| AnvilBench | Coding | 10 of 10 | Nothing to fix | ||
| APEX-1 | Professional work | 11 of 11 | Nothing to fix | ||
| APEX-Accounting | Professional work | 12 of 12 | Nothing to fix | ||
| APEX-Agents | Professional work | 11 of 11 | Nothing to fix | ||
| APEX-Agents 1.1 | Professional work | 10 of 10 | Nothing to fix | ||
| APEX-SWE | Coding | 12 of 12 | Nothing to fix | ||
| API-Bank | Agents and tools | 3 of 3 | Nothing to fix | ||
| APM-Bench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Apollo Research model-welfare research task | Safety and security | 4 of 4 | Nothing to fix | ||
| Apollo Research sabotage evaluations | Safety and security | 3 of 3 | Nothing to fix | ||
| Apple side-by-side human evaluation | Knowledge and language | 5 of 5 | Nothing to fix | ||
| APPS | Coding | 4 of 4 | Nothing to fix | ||
| AppWorld | Agents and tools | 8 of 8 | Nothing to fix | ||
| AppWorld Test-Challenge | Agents and tools | 9 of 9 | Nothing to fix | ||
| AppWorld Test-Normal | Agents and tools | 9 of 9 | Nothing to fix | ||
| AQUA-RAT | Math and science | 7 of 7 | Nothing to fix | ||
| ArabCulture-Dialogue | Knowledge and language | 8 of 8 | Nothing to fix | ||
| AraSentEval 2026 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| AraTrust | Safety and security | 8 of 8 | Nothing to fix | ||
| ARC-AGI | Math and science | 11 of 11 | Nothing to fix | ||
| ARC-AGI-2 | Math and science | 11 of 11 | Nothing to fix | ||
| ARC-Challenge | Knowledge and language | 9 of 9 | Nothing to fix | ||
| ARC-Easy | Knowledge and language | 8 of 8 | Nothing to fix | ||
| ArchDiagram | Coding | 9 of 9 | Nothing to fix | ||
| ARCS | Coding | 10 of 10 | Nothing to fix | ||
| Arena-Hard v2.0 | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Arena-Hard-Auto | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Arena-T2I Hard | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ArenaWebdevCorrectness | Coding | 9 of 9 | Nothing to fix | ||
| ArenaWebdevPreferences | Coding | 9 of 9 | Nothing to fix | ||
| Argo-Bench | Professional work | 10 of 10 | Nothing to fix | ||
| Argus-Bench v2 | Safety and security | 8 of 8 | Nothing to fix | ||
| ARKitScenes | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| ArtEdit-Bench-Lr | Agents and tools | 8 of 8 | Nothing to fix | ||
| Arti4D | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Articulate3D | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ArtifactArena | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| ArtifactsBench | Coding | 11 of 11 | Nothing to fix | ||
| ArtifactsBench (Gemini 2.5 Pro 0506 judge) | Coding | 9 of 9 | Nothing to fix | ||
| Artificial Analysis average word error rate | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| Artificial Analysis Coding Agent Index v1.1 | Coding | 4 of 4 | Nothing to fix | ||
| Artificial Analysis Coding Agent Index v1.4 | Coding | 4 of 4 | Nothing to fix | ||
| Artificial Analysis Coding Agent Index v2 | Coding | 4 of 4 | Nothing to fix | ||
| Artificial Analysis Cyber Index | Safety and security | 5 of 5 | Nothing to fix | ||
| Artificial Turf War | Agents and tools | 9 of 9 | Nothing to fix | ||
| ArXivLean | Math and science | 11 of 11 | Nothing to fix | ||
| ArXivMath | Math and science | 11 of 11 | Nothing to fix | ||
| ASCEND | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| AscendGYM | Coding | 9 of 9 | Nothing to fix | ||
| ASCII Bench (Jason Botterill) | Knowledge and language | 10 of 10 | Nothing to fix | ||
| ASDiv | Math and science | 7 of 7 | Nothing to fix | ||
| ASIMOV v1 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Ask Ubuntu duplicate question retrieval | Knowledge and language | 9 of 9 | Nothing to fix | ||
| ASL-4 virology uplift trial (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| Assembly101-O | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| AssemblyAI Speech Benchmarks | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| AssertLLM2 | Professional work | 8 of 8 | Nothing to fix | ||
| ASSIN 2 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| AssistantBench | Agents and tools | 7 of 7 | Nothing to fix | ||
| AstroCLIMB | Math and science | 8 of 8 | Nothing to fix | ||
| ATEPP | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| ATLAS-Finance | Professional work | 8 of 8 | Nothing to fix | ||
| ATLAS-Finance v1.1.0 | Professional work | 10 of 10 | Nothing to fix | ||
| ATM-Bench | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Atomic Challenges | Safety and security | 9 of 9 | Nothing to fix | ||
| AttaQ | Safety and security | 4 of 4 | Nothing to fix | ||
| Audio Arithmetic: 24 Music Workflow Checks | Math and science | 7 of 7 | Nothing to fix | ||
| Audio Reasoning Challenge 2026 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Audio Understanding Benchmark (Mistral) | Vision, audio and robotics | 3 of 3 | Nothing to fix | ||
| Audio-Visual Event Localization | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| AudioCaps | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| AudioMNIST | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| AudioSep Eval Benchmark | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| AutoBaxBench | Safety and security | 10 of 10 | Nothing to fix | ||
| AutoCAD Bench | Professional work | 10 of 10 | Nothing to fix | ||
| AutoElicit | Safety and security | 9 of 9 | Nothing to fix | ||
| AutoFactory-Bench | Professional work | 11 of 11 | Nothing to fix | ||
| Autoform-Hard | Math and science | 10 of 10 | Nothing to fix | ||
| AutoformBench | Math and science | 10 of 10 | Nothing to fix | ||
| AutoLogi | Math and science | 8 of 8 | Nothing to fix | ||
| Automated Student Assessment Prize: Automated Essay Scoring | Professional work | 8 of 8 | Nothing to fix | ||
| AutoResearchExam | Coding | 10 of 10 | Nothing to fix | ||
| AV-SpeakerBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| AvalonBench | Agents and tools | 7 of 7 | Nothing to fix | ||
| Aviation Blind Spot Benchmark | Professional work | 10 of 10 | Nothing to fix | ||
| AVPS-360 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| AVUT | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| BAAnchmark | Coding | 10 of 10 | Nothing to fix | ||
| BabaIsBench | Agents and tools | 9 of 9 | Nothing to fix | ||
| Babel-ImageNet | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| bAbI QA Tasks | Knowledge and language | 8 of 8 | Nothing to fix | ||
| BabyVision | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Backtracing | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Balatro Bench (Michael Andrzejewski) | Agents and tools | 10 of 10 | Nothing to fix | ||
| BallotBench | Math and science | 10 of 10 | Nothing to fix | ||
| BALROG | Agents and tools | 10 of 10 | Nothing to fix | ||
| BankerToolBench | Professional work | 10 of 10 | Nothing to fix | ||
| BankerToolBench v1.1 | Professional work | 10 of 10 | Nothing to fix | ||
| BANKING77 | Professional work | 9 of 9 | Nothing to fix | ||
| BanSpeech | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| BARC-Bench | Safety and security | 9 of 9 | Nothing to fix | ||
| BAREC Shared Task 2025 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| BAREC-ST-2026 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Bash Agent Evals | Agents and tools | 7 of 7 | Nothing to fix | ||
| Basic reasoning with letters | Knowledge and language | 9 of 9 | Nothing to fix | ||
| BasqueGLUE | Knowledge and language | 7 of 7 | Nothing to fix | ||
| BBEH micro average | Knowledge and language | 11 of 11 | Nothing to fix | ||
| BBQ (Bias Benchmark for QA) | Safety and security | 6 of 6 | Nothing to fix | ||
| BdSLW60 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| BEA 2025 Shared Task on Pedagogical Ability Assessment of AI-powered Tutors | Professional work | 7 of 7 | Nothing to fix | ||
| BEHAVIOR Challenge 2025 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| BEHAVIOR Challenge 2026 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| BEHAVIOR-1K | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Behavior2Code | Coding | 10 of 10 | Nothing to fix | ||
| BEIR FiQA-2018 Retrieval | Professional work | 9 of 9 | Nothing to fix | ||
| Belebele | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Ben-10 Regional ASR | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Bench on the Clocktower | Agents and tools | 10 of 10 | Nothing to fix | ||
| BenchCAD | Professional work | 6 of 6 | Nothing to fix | ||
| BenchCAD 2.0 | Professional work | 11 of 11 | Nothing to fix | ||
| BenchCAD 2.1 | Professional work | 12 of 12 | Nothing to fix | ||
| BenCzechMark | Knowledge and language | 9 of 9 | Nothing to fix | ||
| BetterChartQA | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| Beyond AIME | Math and science | 9 of 9 | Nothing to fix | ||
| bgGLUE | Knowledge and language | 8 of 8 | Nothing to fix | ||
| BhashaBench-Krishi | Knowledge and language | 9 of 9 | Nothing to fix | ||
| BHL OCR Eval | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Big Bench Audio | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| Big Sleep Evaluation (Google) | Safety and security | 10 of 10 | Nothing to fix | ||
| BIG-Bench Hard | Knowledge and language | 9 of 9 | Nothing to fix | ||
| BigCodeBench | Coding | 10 of 10 | Nothing to fix | ||
| BigFinanceBench | Professional work | 10 of 10 | Nothing to fix | ||
| BigLaw Bench | Professional work | 8 of 8 | Nothing to fix | ||
| BigLaw Bench Workflows: SPA Deal Points | Professional work | 7 of 7 | Nothing to fix | ||
| BillSum | Professional work | 8 of 8 | Nothing to fix | ||
| BIM Bench | Professional work | 10 of 10 | Nothing to fix | ||
| BIM-Edit | Professional work | 10 of 10 | Nothing to fix | ||
| Bio Static Jailbreak Eval | Safety and security | 4 of 4 | Nothing to fix | ||
| Bio-PPI Risk: Serous vs Endometrioid | Professional work | 9 of 9 | Nothing to fix | ||
| Bio-Probe | Safety and security | 4 of 4 | Nothing to fix | ||
| Bio-VQA | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| BIOBench (ByteDance Seed) | Safety and security | 4 of 4 | Nothing to fix | ||
| Biochemistry knowledge improvement | Safety and security | 4 of 4 | Nothing to fix | ||
| BioCreative VIII Task 3 Genetic Phenotype Normalization from Dysmorphology Physical Examinations | Professional work | 7 of 7 | Nothing to fix | ||
| Bioinformatics evaluations (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| Biological static hazard evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| Biology Model Refusal Evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| Biology-Instructions | Safety and security | 5 of 5 | Nothing to fix | ||
| Biomedical image analysis (Anthropic) | Math and science | 5 of 5 | Nothing to fix | ||
| BioMysteryBench | Math and science | 11 of 11 | Nothing to fix | ||
| BioNNE 2024 | Professional work | 7 of 7 | Nothing to fix | ||
| BioNNE-L | Professional work | 8 of 8 | Nothing to fix | ||
| BioNNE-R | Professional work | 8 of 8 | Nothing to fix | ||
| BioPhys-Bridge | Safety and security | 9 of 9 | Nothing to fix | ||
| BioPipelineBench (Anthropic) | Math and science | 3 of 3 | Nothing to fix | ||
| BioPipelineBench Verified (Anthropic) | Math and science | 4 of 4 | Nothing to fix | ||
| BioSecBench-Function | Safety and security | 12 of 12 | Nothing to fix | ||
| BioSecBench-Surveillance | Safety and security | 12 of 12 | Nothing to fix | ||
| BioSimBench | Safety and security | 10 of 10 | Nothing to fix | ||
| BioTIER Permit | Safety and security | 11 of 11 | Nothing to fix | ||
| BioTIER Refuse | Safety and security | 11 of 11 | Nothing to fix | ||
| BioTIER v1 Refuse | Safety and security | 4 of 4 | Nothing to fix | ||
| BioVITA retrieval benchmark | Safety and security | 7 of 7 | Nothing to fix | ||
| Bioweapons acquisition uplift trial (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| Bioweapons knowledge questions | Safety and security | 4 of 4 | Nothing to fix | ||
| Bird-SQL | Coding | 11 of 11 | Nothing to fix | ||
| BirdSet | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| BirthdayFacts | Knowledge and language | 3 of 3 | Nothing to fix | ||
| BixBench | Math and science | 9 of 9 | Nothing to fix | ||
| BixBench3 | Math and science | 10 of 10 | Nothing to fix | ||
| BizFinBench | Professional work | 9 of 9 | Nothing to fix | ||
| BLEnD | Knowledge and language | 5 of 5 | Nothing to fix | ||
| BlenderBench (Refresh) | Professional work | 8 of 8 | Nothing to fix | ||
| BlenderWorld | Professional work | 9 of 9 | Nothing to fix | ||
| BLINK | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Blocker Deception | Safety and security | 5 of 5 | Nothing to fix | ||
| Blueprint-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Blueprint-Bench 2 | Vision, audio and robotics | 12 of 12 | Nothing to fix | ||
| BLUEX | Knowledge and language | 8 of 8 | Nothing to fix | ||
| BoardBench 0.1 | Agents and tools | 9 of 9 | Nothing to fix | ||
| BoardBench 0.2 | Agents and tools | 10 of 10 | Nothing to fix | ||
| BoardGameBench (Homer Quan) | Agents and tools | 9 of 9 | Nothing to fix | ||
| BoolQ | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Bornholm Parallel | Knowledge and language | 8 of 8 | Nothing to fix | ||
| BoundaryBench: Policy Under Pressure | Agents and tools | 7 of 7 | Nothing to fix | ||
| BOUQuET | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Brainf**k coding benchmark | Coding | 10 of 10 | Nothing to fix | ||
| Brennan Center election image generation audit | Safety and security | 8 of 8 | Nothing to fix | ||
| Brennan Center election image identification | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| BRIDGE | Professional work | 9 of 9 | Nothing to fix | ||
| Broken Search Tool | Safety and security | 6 of 6 | Nothing to fix | ||
| BrokenArXiv | Math and science | 11 of 11 | Nothing to fix | ||
| Brood War Bench | Agents and tools | 9 of 9 | Nothing to fix | ||
| BRoverbs | Knowledge and language | 8 of 8 | Nothing to fix | ||
| BrowseComp | Agents and tools | 9 of 9 | Nothing to fix | ||
| BrowseComp Extended | Agents and tools | 11 of 11 | Nothing to fix | ||
| BrowseComp Long Context 128k | Knowledge and language | 3 of 3 | Nothing to fix | ||
| BrowseComp-Plus | Agents and tools | 10 of 10 | Nothing to fix | ||
| BrowseComp-VL | Agents and tools | 10 of 10 | Nothing to fix | ||
| Browser Use LinkedIn | Agents and tools | 10 of 10 | Nothing to fix | ||
| BrowseSafe-Adapt | Safety and security | 9 of 9 | Nothing to fix | ||
| BrowseSafe-Bench | Safety and security | 8 of 8 | Nothing to fix | ||
| BrowseSafe++ | Safety and security | 10 of 10 | Nothing to fix | ||
| Browsing prompt injections | Safety and security | 3 of 3 | Nothing to fix | ||
| BRUMO 2025 | Math and science | 8 of 8 | Nothing to fix | ||
| BTF-2 | Math and science | 8 of 8 | Nothing to fix | ||
| BTF-3 | Math and science | 9 of 9 | Nothing to fix | ||
| BU Bench V2 | Agents and tools | 10 of 10 | Nothing to fix | ||
| BU Bench V2.1 | Agents and tools | 10 of 10 | Nothing to fix | ||
| Build-A-Chip-Bench | Professional work | 10 of 10 | Nothing to fix | ||
| BuildArena | Agents and tools | 9 of 9 | Nothing to fix | ||
| BuildArena 2.0 Construction Challenge | Agents and tools | 8 of 8 | Nothing to fix | ||
| BuilderBench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| BullshitBench | Knowledge and language | 10 of 10 | Nothing to fix | ||
| BullshitBench V2 | Knowledge and language | 10 of 10 | Nothing to fix | ||
| BURST | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| BuseyBench SVG Test | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Business Evidence Judgment v4 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| BVA Case Citation Dataset | Professional work | 7 of 7 | Nothing to fix | ||
| BVB | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| C-Eval | Knowledge and language | 9 of 9 | Nothing to fix | ||
| C3 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| CACAE 2025 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| CAD Arena | Professional work | 10 of 10 | Nothing to fix | ||
| CADClamp 0.1 | Professional work | 9 of 9 | Nothing to fix | ||
| CADClamp 0.2 | Professional work | 10 of 10 | Nothing to fix | ||
| CADECv2 | Professional work | 7 of 7 | Nothing to fix | ||
| CADPrompt | Professional work | 7 of 7 | Nothing to fix | ||
| CAIS Vision Capabilities Index | Other | 10 of 10 | Nothing to fix | ||
| CALVIN | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| CAMeL | Safety and security | 8 of 8 | Nothing to fix | ||
| Can AI close the books? FP&A reasoning | Professional work | 10 of 10 | Nothing to fix | ||
| Can I Play It? | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| CanItEdit | Coding | 4 of 4 | Nothing to fix | ||
| CAPR-am | Knowledge and language | 9 of 9 | Nothing to fix | ||
| CAPR-ar | Knowledge and language | 9 of 9 | Nothing to fix | ||
| CAPR-ja | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Capture the Flag (Collegiate) | Safety and security | 5 of 5 | Nothing to fix | ||
| Capture the Flag (High School) | Safety and security | 5 of 5 | Nothing to fix | ||
| Capture the Flag (Internal) | Safety and security | 4 of 4 | Nothing to fix | ||
| Capture the Flag (Professional) | Safety and security | 4 of 4 | Nothing to fix | ||
| CAR-bench | Agents and tools | 10 of 10 | Nothing to fix | ||
| CAREBench | Safety and security | 9 of 9 | Nothing to fix | ||
| CASE-Bench | Safety and security | 8 of 8 | Nothing to fix | ||
| Cat-Goric Game Benchmark | Agents and tools | 9 of 9 | Nothing to fix | ||
| Catalonia Independence Corpus | Knowledge and language | 7 of 7 | Nothing to fix | ||
| CATArena Pommerman | Coding | 7 of 7 | Nothing to fix | ||
| CATArena Tetris | Coding | 7 of 7 | Nothing to fix | ||
| CauGym | Knowledge and language | 7 of 7 | Nothing to fix | ||
| CB | Knowledge and language | 8 of 8 | Nothing to fix | ||
| CBRN Conversational Evals | Safety and security | 4 of 4 | Nothing to fix | ||
| CBRN Science Eval | Safety and security | 4 of 4 | Nothing to fix | ||
| CBT-Bench Level II | Professional work | 10 of 10 | Nothing to fix | ||
| CC-OCR | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| CC-OCR-Bench-V2 | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| CCBench (MMBench Chinese Culture) | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| CCPM | Knowledge and language | 3 of 3 | Nothing to fix | ||
| CDVQA | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Cekura Agent Performance | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Cekura Production Voice Agents | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Cekura Real Human Calls | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Cekura Text to Speech | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Cekura Turn Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Cents Matter | Professional work | 8 of 8 | Nothing to fix | ||
| CEO-Bench | Agents and tools | 9 of 9 | Nothing to fix | ||
| CEQuest | Professional work | 10 of 10 | Nothing to fix | ||
| CF-Div2-Stepfun | Coding | 9 of 9 | Nothing to fix | ||
| CFDCodeBench | Coding | 9 of 9 | Nothing to fix | ||
| CFDCodeBench V2 | Coding | 10 of 10 | Nothing to fix | ||
| CFDQuery | Math and science | 9 of 9 | Nothing to fix | ||
| Chain-of-thought comprehensibility | Safety and security | 4 of 4 | Nothing to fix | ||
| Chain-of-thought informativeness | Safety and security | 4 of 4 | Nothing to fix | ||
| Challenging prompts from red teamers with biosafety-relevant PhDs | Safety and security | 3 of 3 | Nothing to fix | ||
| Challenging Red Teaming Evaluation 1 (created for o3-mini) | Safety and security | 4 of 4 | Nothing to fix | ||
| Challenging Red Teaming Evaluation 2 (created for deep research) | Safety and security | 4 of 4 | Nothing to fix | ||
| Challenging Refusal Evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| Challenging translation human evaluation (Z.ai) | Knowledge and language | 3 of 3 | Nothing to fix | ||
| ChangeMyView | Safety and security | 4 of 4 | Nothing to fix | ||
| CHAOS-Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Charlemagne Labs social-engineering benchmark | Safety and security | 4 of 4 | Nothing to fix | ||
| Chart Arena | Coding | 10 of 10 | Nothing to fix | ||
| ChartQA | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| ChartQA Pro | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| CharXiv | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| CharXiv Extended | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| CharXiv Missing Image | Safety and security | 3 of 3 | Nothing to fix | ||
| CharXiv Missing Images | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Chatbots and the Ballot Box | Knowledge and language | 8 of 8 | Nothing to fix | ||
| ChCatExt | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Chem / Bio Dynamic Jailbreak Eval | Safety and security | 4 of 4 | Nothing to fix | ||
| ChemDIRT | Math and science | 10 of 10 | Nothing to fix | ||
| ChemicalDiagramQA | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| ChemProt | Safety and security | 8 of 8 | Nothing to fix | ||
| Chess Engine Benchmark (Steve Maughan) | Coding | 10 of 10 | Nothing to fix | ||
| ChessBench (MadeByLupu) | Agents and tools | 10 of 10 | Nothing to fix | ||
| ChestX-ray14 | Professional work | 7 of 7 | Nothing to fix | ||
| CheXpert | Professional work | 8 of 8 | Nothing to fix | ||
| CHI-Bench | Professional work | 10 of 10 | Nothing to fix | ||
| Child Safety Action Testing | Safety and security | 4 of 4 | Nothing to fix | ||
| Child Sexual Abuse and Exploitation Safety | Safety and security | 4 of 4 | Nothing to fix | ||
| CHiME-6 Track 2: Unsegmented Speech Recognition | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ChinaBench | Safety and security | 10 of 10 | Nothing to fix | ||
| Chinese Classical Poetry Appreciation 2025 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Chinese Complex Instruction Following (ByteDance Seed in-house) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Chinese Essay Discourse Coherence Corpus | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Chinese Essay Fluency Evaluation 2023 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Chinese Essay Fluency Evaluation 2024 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Chinese Essay Rhetoric Recognition and Understanding 2024 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Chinese Essay Rhetoric Recognition Evaluation 2025 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Chinese Language Generation Evaluation (CLGE) | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Chinese Machine Reading Comprehension 2017 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Chinese Machine Reading Comprehension 2019 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Chinese SimpleQA | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Chinese Student English Essay | Professional work | 7 of 7 | Nothing to fix | ||
| Chinese Tool-Use Benchmark | Agents and tools | 3 of 3 | Nothing to fix | ||
| Chinese WPLC | Knowledge and language | 8 of 8 | Nothing to fix | ||
| ChineseQA | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Chrome Production Commit Scanning (Google) | Safety and security | 10 of 10 | Nothing to fix | ||
| ChronoMagic-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Chronos Benchmark II | Math and science | 8 of 8 | Nothing to fix | ||
| CineRecon | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| CircuitRubric | Professional work | 9 of 9 | Nothing to fix | ||
| CIRR | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Cityscapes | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| CivilComments-WILDS | Safety and security | 7 of 7 | Nothing to fix | ||
| CL-bench | Knowledge and language | 11 of 11 | Nothing to fix | ||
| CL-bench Life | Knowledge and language | 11 of 11 | Nothing to fix | ||
| CLARA-MeD Spanish Medical Text Simplification | Professional work | 8 of 8 | Nothing to fix | ||
| Clash of AIs Season 1 | Professional work | 10 of 10 | Nothing to fix | ||
| CLASHarness | Agents and tools | 10 of 10 | Nothing to fix | ||
| ClassEval | Coding | 8 of 8 | Nothing to fix | ||
| Classical Chinese Historical Event Detection | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Claw-Eval Pass^3 | Agents and tools | 11 of 11 | Nothing to fix | ||
| ClawBench | Agents and tools | 9 of 9 | Nothing to fix | ||
| CLEAR French Medical Text Simplification | Professional work | 8 of 8 | Nothing to fix | ||
| CLEF 2025 SimpleText Task 2 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| CLEF 2026 SimpleText Task 1 | Professional work | 8 of 8 | Nothing to fix | ||
| CLEF HIPE 2026 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| CLEF JOKER 2025 Task 1: Humour-Aware Information Retrieval | Knowledge and language | 7 of 7 | Nothing to fix | ||
| CLEF JOKER 2025 Task 2: Wordplay Translation | Knowledge and language | 8 of 8 | Nothing to fix | ||
| CLEF JOKER 2025 Task 3: Onomastic Wordplay Translation | Knowledge and language | 8 of 8 | Nothing to fix | ||
| CLEF JOKER 2026 Task 1: Humour-Aware Information Retrieval | Knowledge and language | 7 of 7 | Nothing to fix | ||
| CLEF JOKER 2026 Task 2: Pun Translation | Knowledge and language | 8 of 8 | Nothing to fix | ||
| clembench | Agents and tools | 9 of 9 | Nothing to fix | ||
| CLI-Bench One-Shot | Coding | 9 of 9 | Nothing to fix | ||
| CLIcK | Knowledge and language | 4 of 4 | Nothing to fix | ||
| CLICK-ID | Knowledge and language | 10 of 10 | Nothing to fix | ||
| CLIF Bench | Knowledge and language | 5 of 5 | Nothing to fix | ||
| ClimateCheck 2025 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| ClimateCheck 2026 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| CliniCARE-Bench | Professional work | 10 of 10 | Nothing to fix | ||
| CLIP representation bias evaluation | Safety and security | 7 of 7 | Nothing to fix | ||
| Clipping Bench | Professional work | 10 of 10 | Nothing to fix | ||
| CLIPSep Sound Separation | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ClockBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Clocktower Bench | Agents and tools | 10 of 10 | Nothing to fix | ||
| Closed-Loop Evaluation of LLM Agents for Embedded Software Development | Professional work | 8 of 8 | Nothing to fix | ||
| Clotho | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Clotho-AQA | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| CLUE | Knowledge and language | 8 of 8 | Nothing to fix | ||
| CLUENER2020 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| CLUEWSC | Knowledge and language | 8 of 8 | Nothing to fix | ||
| ClutterBench | Coding | 10 of 10 | Nothing to fix | ||
| CMATH | Math and science | 4 of 4 | Nothing to fix | ||
| CMD-AD-Eval | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| CMIMC 2025 | Math and science | 8 of 8 | Nothing to fix | ||
| CMMLU | Knowledge and language | 10 of 10 | Nothing to fix | ||
| CNFinBench | Professional work | 9 of 9 | Nothing to fix | ||
| Co-Design Bench | Coding | 10 of 10 | Nothing to fix | ||
| Cochrane RCT Summarization | Professional work | 8 of 8 | Nothing to fix | ||
| Cochrane Simplification | Professional work | 7 of 7 | Nothing to fix | ||
| Cochrane-auto | Professional work | 8 of 8 | Nothing to fix | ||
| COCO | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| COCO-O | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| COCO-Stuff 164K | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Code Migration | Coding | 11 of 11 | Nothing to fix | ||
| Code Migration v1.1 | Coding | 10 of 10 | Nothing to fix | ||
| Code Review Bench (Martian) | Coding | 8 of 8 | Nothing to fix | ||
| Code summary honesty (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| Code2Video Arena | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Code2Video Benchmark | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Code2Video-AA | Coding | 4 of 4 | Nothing to fix | ||
| Code2Video-Compass | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Code4Scene | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| CodeEditorBench | Coding | 8 of 8 | Nothing to fix | ||
| CodeIPI | Safety and security | 8 of 8 | Nothing to fix | ||
| CodeSearchNet Challenge | Coding | 9 of 9 | Nothing to fix | ||
| CodeTransOcean | Coding | 9 of 9 | Nothing to fix | ||
| CodeUtilityBench (LG) | Coding | 4 of 4 | Nothing to fix | ||
| codex of veritas | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Coding (MiniMax in-house) | Coding | 3 of 3 | Nothing to fix | ||
| Coding agent prompt injection robustness | Safety and security | 5 of 5 | Nothing to fix | ||
| Coding Deception | Safety and security | 5 of 5 | Nothing to fix | ||
| Coding prompt injections | Safety and security | 3 of 3 | Nothing to fix | ||
| COIN instructional video analysis | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| CoIR | Coding | 8 of 8 | Nothing to fix | ||
| ColDeR | Knowledge and language | 8 of 8 | Nothing to fix | ||
| COLIEE 2025 | Professional work | 8 of 8 | Nothing to fix | ||
| COLIEE 2026 | Professional work | 8 of 8 | Nothing to fix | ||
| Collab-Overcooked | Agents and tools | 8 of 8 | Nothing to fix | ||
| Collaborative Intelligence Benchmark | Agents and tools | 11 of 11 | Nothing to fix | ||
| COLLIE | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Combined HHH evals (Anthropic) | Safety and security | 8 of 8 | Nothing to fix | ||
| Command Code Flappy Bird | Coding | 9 of 9 | Nothing to fix | ||
| Command Code Pixel Art Space Shooter | Coding | 9 of 9 | Nothing to fix | ||
| CommerceAgentBench | Professional work | 8 of 8 | Nothing to fix | ||
| CommonSenseQA | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Company Agent Bench | Professional work | 11 of 11 | Nothing to fix | ||
| Company Enrichment API Benchmark | Agents and tools | 8 of 8 | Nothing to fix | ||
| Competitive Programming Code Generation | Coding | 9 of 9 | Nothing to fix | ||
| ComplaintsClustering | Professional work | 8 of 8 | Nothing to fix | ||
| Complex Workflow (ByteDance Seed in-house) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| ComplexConstraints | Knowledge and language | 12 of 12 | Nothing to fix | ||
| ComplexFuncBench Audio (OpenAI) | Agents and tools | 4 of 4 | Nothing to fix | ||
| Compliance with instructions planted in pasted text (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| CompMat-Bench | Math and science | 8 of 8 | Nothing to fix | ||
| Composite-Bench | Agents and tools | 10 of 10 | Nothing to fix | ||
| Computer use manual red teaming | Safety and security | 4 of 4 | Nothing to fix | ||
| Computer use Prohibited Action Evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| Computer use prompt injection evaluation (Anthropic) | Safety and security | 5 of 5 | Nothing to fix | ||
| Computer use prompt injection robustness | Safety and security | 4 of 4 | Nothing to fix | ||
| Concealment of changes (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| Construction AI Leaderboard | Other | 10 of 10 | Nothing to fix | ||
| Construction's Last Exam | Professional work | 11 of 11 | Nothing to fix | ||
| Consulting and Recommendations Benchmark (StepFun) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Context Aware Automatic Speech Recognition | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| Contextual Nuclear Knowledge | Safety and security | 4 of 4 | Nothing to fix | ||
| Contextual security evaluation | Safety and security | 5 of 5 | Nothing to fix | ||
| Continual Learning Exam | Knowledge and language | 10 of 10 | Nothing to fix | ||
| ContractNLI | Professional work | 8 of 8 | Nothing to fix | ||
| ContractorOS Quantity Takeoff | Professional work | 11 of 11 | Nothing to fix | ||
| COPA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Coronavirus-ACE2 Cell-Entry Screen | Safety and security | 6 of 6 | Nothing to fix | ||
| Corp Speak Arena | Knowledge and language | 10 of 10 | Nothing to fix | ||
| CorpFin v2 | Professional work | 11 of 11 | Nothing to fix | ||
| CorporateBench | Knowledge and language | 11 of 11 | Nothing to fix | ||
| Corpus for Emotion Detection in Russian | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Corpus of Multilingual Code-switched Soap Opera Speech | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| CorpusQA | Knowledge and language | 5 of 5 | Nothing to fix | ||
| Corr2Cause | Knowledge and language | 8 of 8 | Nothing to fix | ||
| CoSQA | Coding | 9 of 9 | Nothing to fix | ||
| Cost Per Correct v2 | Agents and tools | 8 of 8 | Nothing to fix | ||
| Cost per Successful Task | Agents and tools | 9 of 9 | Nothing to fix | ||
| CoT Deception Monitoring | Safety and security | 3 of 3 | Nothing to fix | ||
| CoT-Control | Safety and security | 7 of 7 | Nothing to fix | ||
| CounselBench-Adv | Safety and security | 10 of 10 | Nothing to fix | ||
| CounselBench-Eval | Professional work | 10 of 10 | Nothing to fix | ||
| CountBench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| CounterBench | Knowledge and language | 8 of 8 | Nothing to fix | ||
| CounterBench (1,200 questions) | Knowledge and language | 8 of 8 | Nothing to fix | ||
| CountQA | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Country211 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Coval Speech-to-Text WER | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Covert misbehavior (Thinking Machines) | Safety and security | 4 of 4 | Nothing to fix | ||
| CoVoST2 | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| CoWorkBench | Professional work | 11 of 11 | Nothing to fix | ||
| CPRet | Coding | 7 of 7 | Nothing to fix | ||
| CPU Inference Optimization and Streaming | Coding | 8 of 8 | Nothing to fix | ||
| CREATE | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Creative biology | Safety and security | 5 of 5 | Nothing to fix | ||
| Creative Biology Uplift Trial (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| Creative Control: LLMs as a Film Crew | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Creative Writing (MiniMax in-house) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| CreativeWork | Agents and tools | 11 of 11 | Nothing to fix | ||
| CredData | Safety and security | 7 of 7 | Nothing to fix | ||
| Credit Firewall Benchmark | Safety and security | 9 of 9 | Nothing to fix | ||
| CREF | Math and science | 8 of 8 | Nothing to fix | ||
| Crest Arcade Chess | Agents and tools | 10 of 10 | Nothing to fix | ||
| Crest Arcade Elemental Armory | Agents and tools | 11 of 11 | Nothing to fix | ||
| Crest Arcade FreeCell | Agents and tools | 11 of 11 | Nothing to fix | ||
| Crest Arcade Hustle Sim | Agents and tools | 10 of 10 | Nothing to fix | ||
| Crest Arcade Micro Mouse | Agents and tools | 10 of 10 | Nothing to fix | ||
| Crest Arcade Mines | Agents and tools | 10 of 10 | Nothing to fix | ||
| Crest Arcade Nations Royale | Agents and tools | 10 of 10 | Nothing to fix | ||
| Crest Arcade Nonograms | Agents and tools | 10 of 10 | Nothing to fix | ||
| Crest Arcade Ravelin | Agents and tools | 10 of 10 | Nothing to fix | ||
| Crest Arcade Starjar | Agents and tools | 10 of 10 | Nothing to fix | ||
| Crest Arcade Sudoku | Agents and tools | 10 of 10 | Nothing to fix | ||
| Crest Arcade Wildhaven | Agents and tools | 10 of 10 | Nothing to fix | ||
| CrisisMMD | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| CritPt | Math and science | 11 of 11 | Nothing to fix | ||
| CritPt v2 | Math and science | 11 of 11 | Nothing to fix | ||
| CRM-Agent-Bench | Professional work | 10 of 10 | Nothing to fix | ||
| Cross-lingual Test Dataset | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Crossmodal-3600 | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| CrossSum | Knowledge and language | 8 of 8 | Nothing to fix | ||
| CrossVid | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| CRSBench | Safety and security | 10 of 10 | Nothing to fix | ||
| CRUXEval | Coding | 8 of 8 | Nothing to fix | ||
| CryptanalysisBench | Safety and security | 10 of 10 | Nothing to fix | ||
| Crypto CTF challenges (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| CSL | Knowledge and language | 7 of 7 | Nothing to fix | ||
| CT-RATE | Professional work | 7 of 7 | Nothing to fix | ||
| CTF-Archive Diamond | Safety and security | 10 of 10 | Nothing to fix | ||
| CTI-to-MITRE | Safety and security | 8 of 8 | Nothing to fix | ||
| CTKFacts | Knowledge and language | 9 of 9 | Nothing to fix | ||
| CUA-bench | Agents and tools | 11 of 11 | Nothing to fix | ||
| CUAD | Professional work | 8 of 8 | Nothing to fix | ||
| CueSpeak | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| CulturalRiddles | Knowledge and language | 10 of 10 | Nothing to fix | ||
| CURE | Professional work | 8 of 8 | Nothing to fix | ||
| CursorBench 3.1 | Coding | 11 of 11 | Nothing to fix | ||
| CursorBench 4.0 | Coding | 12 of 12 | Nothing to fix | ||
| Customer Support Q&A (ByteDance Seed in-house) | Agents and tools | 4 of 4 | Nothing to fix | ||
| CVBench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| CVDP | Professional work | 11 of 11 | Nothing to fix | ||
| CVE Rediscovery (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| CVE-Bench | Safety and security | 9 of 9 | Nothing to fix | ||
| CWE-bench | Safety and security | 12 of 12 | Nothing to fix | ||
| CWE-bench 2 | Safety and security | 12 of 12 | Nothing to fix | ||
| Cybench | Safety and security | 9 of 9 | Nothing to fix | ||
| Cyber Adaptive Jailbreak Eval | Safety and security | 4 of 4 | Nothing to fix | ||
| Cyber agentic safeguard evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| Cyber Conversational Evals | Safety and security | 4 of 4 | Nothing to fix | ||
| Cyber ECI | Other | 10 of 10 | Nothing to fix | ||
| Cyber harm coverage (Anthropic) | Safety and security | 6 of 6 | Nothing to fix | ||
| Cyber key skills benchmark v1 | Safety and security | 3 of 3 | Nothing to fix | ||
| Cyber key skills benchmark v2 | Safety and security | 3 of 3 | Nothing to fix | ||
| Cyber Range | Safety and security | 5 of 5 | Nothing to fix | ||
| Cyber Static Jailbreak Eval | Safety and security | 4 of 4 | Nothing to fix | ||
| Cyber Utility | Safety and security | 5 of 5 | Nothing to fix | ||
| Cyber-harness network challenges (Incalmo cyber ranges) | Safety and security | 4 of 4 | Nothing to fix | ||
| CyberBench (Vals) | Safety and security | 9 of 9 | Nothing to fix | ||
| CyberBench v1.1 (Vals) | Safety and security | 12 of 12 | Nothing to fix | ||
| CyberGym | Safety and security | 10 of 10 | Nothing to fix | ||
| CyberGym-E2E | Safety and security | 9 of 9 | Nothing to fix | ||
| CyberOffense | Safety and security | 9 of 9 | Nothing to fix | ||
| CyberSecEval Autocomplete | Safety and security | 10 of 10 | Nothing to fix | ||
| CyberSecEval Code Interpreter Abuse | Safety and security | 9 of 9 | Nothing to fix | ||
| CyberSecEval Instruct | Safety and security | 10 of 10 | Nothing to fix | ||
| CyberSOCEval Malware Analysis | Safety and security | 10 of 10 | Nothing to fix | ||
| CyberSOCEval Threat Intelligence Reasoning | Safety and security | 5 of 5 | Nothing to fix | ||
| CyberStrike | Safety and security | 10 of 10 | Nothing to fix | ||
| DABstep | Coding | 8 of 8 | Nothing to fix | ||
| DACSA | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Daily-Omni | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| DaLAJ | Knowledge and language | 8 of 8 | Nothing to fix | ||
| DaLAJ-GED 2.0 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Daleel 2026 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Dan’s Editorial Checks | Knowledge and language | 10 of 10 | Nothing to fix | ||
| DaNE | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Dangerous Content policy violations (from ART) | Safety and security | 3 of 3 | Nothing to fix | ||
| Dangerous Content Safety | Safety and security | 4 of 4 | Nothing to fix | ||
| Danish Semantic Reasoning Benchmark | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Data Agent Benchmark | Coding | 10 of 10 | Nothing to fix | ||
| Data Analysis Benchmark (StepFun) | Coding | 4 of 4 | Nothing to fix | ||
| DataAnalyticsBench | Professional work | 10 of 10 | Nothing to fix | ||
| DataBench v1.1 | Professional work | 11 of 11 | Nothing to fix | ||
| DataCenterBench Tier 1 | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| DataComp image encoder evaluation | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Dấu or Die: Vietnamese Diacritic Restoration | Knowledge and language | 9 of 9 | Nothing to fix | ||
| DAVIS 2017 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DAYJOB: Finance | Professional work | 12 of 12 | Nothing to fix | ||
| DAYJOB: Healthcare | Professional work | 11 of 11 | Nothing to fix | ||
| DAYJOB: Tax | Professional work | 11 of 11 | Nothing to fix | ||
| DBpedia 14 | Knowledge and language | 9 of 9 | Nothing to fix | ||
| DCASE 2023 Foley Sound Synthesis | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DCASE 2024 Language-Queried Audio Source Separation | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DCASE 2024 Sound Scene Synthesis | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DCS 2018 Sanskrit Word Segmentation | Knowledge and language | 7 of 7 | Nothing to fix | ||
| DDisCo | Knowledge and language | 7 of 7 | Nothing to fix | ||
| DDXPlus | Professional work | 7 of 7 | Nothing to fix | ||
| De novo protein binder design (Anthropic) | Math and science | 5 of 5 | Nothing to fix | ||
| DecBench | Coding | 9 of 9 | Nothing to fix | ||
| Deception in production traffic | Safety and security | 3 of 3 | Nothing to fix | ||
| DECIDE | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Decision Index | Other | 9 of 9 | Nothing to fix | ||
| Decision Index Public v0.3 | Other | 4 of 4 | Nothing to fix | ||
| Decision Model Benchmark | Knowledge and language | 9 of 9 | Nothing to fix | ||
| DECK-Arena | Professional work | 4 of 4 | Nothing to fix | ||
| Declining to Exploit a Honeypot During Difficult ExploitGym Problems | Safety and security | 6 of 6 | Nothing to fix | ||
| Decodability Assessment Benchmark | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Deep Research Bench (Moonshot) | Agents and tools | 4 of 4 | Nothing to fix | ||
| Deep20Bench 1.1 | Agents and tools | 10 of 10 | Nothing to fix | ||
| DeepFinance | Professional work | 11 of 11 | Nothing to fix | ||
| Deepgram Voice AI Benchmarks | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DeepPlanning | Agents and tools | 10 of 10 | Nothing to fix | ||
| DeepResearchEval | Agents and tools | 8 of 8 | Nothing to fix | ||
| DeepResearchEval v1 | Agents and tools | 8 of 8 | Nothing to fix | ||
| DeepRSI | Coding | 10 of 10 | Nothing to fix | ||
| DeepSearchQA | Agents and tools | 9 of 9 | Nothing to fix | ||
| DeepsecBench | Safety and security | 10 of 10 | Nothing to fix | ||
| DeepSeek Chinese white-collar tasks evaluation | Professional work | 4 of 4 | Nothing to fix | ||
| DeepSeek in-house safety benchmark | Safety and security | 3 of 3 | Nothing to fix | ||
| DeepSeek jailbreak robustness evaluation | Safety and security | 3 of 3 | Nothing to fix | ||
| DeepSeek multilingual safety evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| DeepSeek R&D Coding Benchmark | Coding | 4 of 4 | Nothing to fix | ||
| DeepSWE | Coding | 12 of 12 | Nothing to fix | ||
| Default safety evaluation (Cohere) | Safety and security | 4 of 4 | Nothing to fix | ||
| DELTA | Professional work | 9 of 9 | Nothing to fix | ||
| Delusional Reinforcement evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| dependency-bench | Coding | 8 of 8 | Nothing to fix | ||
| DeployGuard | Coding | 8 of 8 | Nothing to fix | ||
| DeputyBench | Professional work | 10 of 10 | Nothing to fix | ||
| Describable Textures Dataset | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DESED | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Design Bench (Moonshot) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| DesignArena 3D Design | Coding | 11 of 11 | Nothing to fix | ||
| DesignArena Agentic Game Dev | Coding | 10 of 10 | Nothing to fix | ||
| DesignArena Agentic Overall | Coding | 10 of 10 | Nothing to fix | ||
| DesignArena Android (Kotlin) | Coding | 10 of 10 | Nothing to fix | ||
| DesignArena ASCII Art | Knowledge and language | 10 of 10 | Nothing to fix | ||
| DesignArena Audio Realism Bench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| DesignArena Blender | Coding | 10 of 10 | Nothing to fix | ||
| DesignArena Builder | Coding | 8 of 8 | Nothing to fix | ||
| DesignArena Code | Coding | 10 of 10 | Nothing to fix | ||
| DesignArena Code to Video | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| DesignArena Data Visualization | Coding | 10 of 10 | Nothing to fix | ||
| DesignArena Frontend Web App | Coding | 11 of 11 | Nothing to fix | ||
| DesignArena Full Stack | Coding | 10 of 10 | Nothing to fix | ||
| DesignArena Game Dev | Coding | 11 of 11 | Nothing to fix | ||
| DesignArena Godot Game Dev | Coding | 10 of 10 | Nothing to fix | ||
| DesignArena HTML Slides | Professional work | 10 of 10 | Nothing to fix | ||
| DesignArena Image | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| DesignArena Image Editing | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| DesignArena Image to Frontend | Coding | 10 of 10 | Nothing to fix | ||
| DesignArena Image to HTML | Coding | 10 of 10 | Nothing to fix | ||
| DesignArena Image to Video | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| DesignArena Mobile Builder | Coding | 8 of 8 | Nothing to fix | ||
| DesignArena Multi-Image to Video | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| DesignArena Multimodal to Video | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| DesignArena PDF | Professional work | 10 of 10 | Nothing to fix | ||
| DesignArena Proprietary Harness Slides | Professional work | 8 of 8 | Nothing to fix | ||
| DesignArena Python-PPTX Slides | Professional work | 10 of 10 | Nothing to fix | ||
| DesignArena React Native | Coding | 10 of 10 | Nothing to fix | ||
| DesignArena Spreadsheet | Professional work | 10 of 10 | Nothing to fix | ||
| DesignArena SVG | Coding | 10 of 10 | Nothing to fix | ||
| DesignArena Text-to-Speech | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| DesignArena UI Component | Coding | 10 of 10 | Nothing to fix | ||
| DesignArena Video Editing | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DesignArena Video to Website | Coding | 9 of 9 | Nothing to fix | ||
| DesignArena Website | Coding | 11 of 11 | Nothing to fix | ||
| Destructive action avoidance | Safety and security | 5 of 5 | Nothing to fix | ||
| Destructive actions (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| Destructive actions (Thinking Machines) | Safety and security | 4 of 4 | Nothing to fix | ||
| Destructive production evaluation (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| Detecting Entities in the Astrophysics Literature | Math and science | 8 of 8 | Nothing to fix | ||
| Deterministic Agent Reasoning & Formatting Suite | Knowledge and language | 8 of 8 | Nothing to fix | ||
| DETESTS-Dis 2024 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Developer Retrieval Benchmark | Agents and tools | 8 of 8 | Nothing to fix | ||
| DEW-LogiQ | Math and science | 7 of 7 | Nothing to fix | ||
| dfbench | Safety and security | 10 of 10 | Nothing to fix | ||
| DFKI English-Russian Linguistic Test Suite at WMT25 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| DiaBLa | Knowledge and language | 9 of 9 | Nothing to fix | ||
| DialectSentEval 2026 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Dictatorship Eval | Safety and security | 9 of 9 | Nothing to fix | ||
| DiDeMo | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Digital Work Index | Professional work | 10 of 10 | Nothing to fix | ||
| DIHARD III | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DiligenceBench | Professional work | 5 of 5 | Nothing to fix | ||
| DIODE | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DIOR | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DIOR-RSVG | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Dirty Integration | Agents and tools | 7 of 7 | Nothing to fix | ||
| DISCFinLLMReranking | Professional work | 8 of 8 | Nothing to fix | ||
| DiSCoMaT | Math and science | 8 of 8 | Nothing to fix | ||
| DISPLACE Challenge 2024 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| DISPLACE-M 2026 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| DisTEMIST | Professional work | 7 of 7 | Nothing to fix | ||
| DistressBench | Safety and security | 11 of 11 | Nothing to fix | ||
| Divergence attack (extractable memorization) | Safety and security | 3 of 3 | Nothing to fix | ||
| Diving48 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DNA sequence design for transcription factor binding | Safety and security | 5 of 5 | Nothing to fix | ||
| DNA Synthesis Screening Evasion | Safety and security | 6 of 6 | Nothing to fix | ||
| Do-Not-Answer | Safety and security | 8 of 8 | Nothing to fix | ||
| Doc2Harness | Coding | 10 of 10 | Nothing to fix | ||
| Document AI Benchmark Arena | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DolphinBench | Agents and tools | 9 of 9 | Nothing to fix | ||
| Dosimeter AI Disinformation Benchmark | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Dosimeter US midterm source audit | Knowledge and language | 8 of 8 | Nothing to fix | ||
| doteBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Doubao Multi-Turn Bench | Knowledge and language | 11 of 11 | Nothing to fix | ||
| DRACO | Agents and tools | 10 of 10 | Nothing to fix | ||
| DRAMA-Eval | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DravidianLangTech 2026 Multilevel Political Meme Classification in Tamil and Malayalam | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Dreaddit | Safety and security | 8 of 8 | Nothing to fix | ||
| DreamGen Bench | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| DrivingBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| DrivingBench v2 | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| DroidTask | Agents and tools | 7 of 7 | Nothing to fix | ||
| Drone-Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| DROP | Knowledge and language | 9 of 9 | Nothing to fix | ||
| DrugDiscoveryBench | Safety and security | 11 of 11 | Nothing to fix | ||
| DrugTargetWorld | Safety and security | 9 of 9 | Nothing to fix | ||
| DSBench | Coding | 3 of 3 | Nothing to fix | ||
| DSBench (DeepSeek internal) | Coding | 5 of 5 | Nothing to fix | ||
| DSI-Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| DTap Direct Misuse | Safety and security | 9 of 9 | Nothing to fix | ||
| DTap Prompt Injection | Safety and security | 9 of 9 | Nothing to fix | ||
| DTBench | Math and science | 10 of 10 | Nothing to fix | ||
| Dual-Arm Sim | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Dubesor AI Chess Leaderboard | Agents and tools | 10 of 10 | Nothing to fix | ||
| Dubesor Lichess Puzzle Rating | Agents and tools | 10 of 10 | Nothing to fix | ||
| DuetBench | Agents and tools | 9 of 9 | Nothing to fix | ||
| Dutch Book Bench | Math and science | 9 of 9 | Nothing to fix | ||
| Dutch Book Reviews Dataset | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Dutch Model Benchmark | Knowledge and language | 8 of 8 | Nothing to fix | ||
| DynaMath | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| E-Bench (Tencent) | Professional work | 11 of 11 | Nothing to fix | ||
| E-Bench-Code (Tencent) | Professional work | 11 of 11 | Nothing to fix | ||
| E-Commerce Bench | Professional work | 9 of 9 | Nothing to fix | ||
| E/RP Leaderboard by PsychoSmiley | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Earnings-21 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Earnings-22 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Earth-Bench | Agents and tools | 9 of 9 | Nothing to fix | ||
| Earth-Bench V2 | Agents and tools | 11 of 11 | Nothing to fix | ||
| Earth-Bench-Pro | Agents and tools | 8 of 8 | Nothing to fix | ||
| EarthRecord | Math and science | 10 of 10 | Nothing to fix | ||
| EBench | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| ECHR Binary Violation Classification | Professional work | 8 of 8 | Nothing to fix | ||
| ECHR Legal Corpus for Argument Mining | Professional work | 7 of 7 | Nothing to fix | ||
| EDIS | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Education (ByteDance Seed in-house) | Professional work | 4 of 4 | Nothing to fix | ||
| EduExplain | Professional work | 7 of 7 | Nothing to fix | ||
| EduVideoBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Ego2Flow | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Ego4D Goal-Step | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Ego4D State Change Object Detection | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| EgoCross | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| EgoHome | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| EgoProceL | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| EHRNoteQA | Professional work | 8 of 8 | Nothing to fix | ||
| EHRSQL | Coding | 7 of 7 | Nothing to fix | ||
| EJBench | Math and science | 10 of 10 | Nothing to fix | ||
| EKWbench | Professional work | 10 of 10 | Nothing to fix | ||
| ELCardioCC 2025 | Professional work | 7 of 7 | Nothing to fix | ||
| Election-Day AI Poisoning in Germany | Knowledge and language | 8 of 8 | Nothing to fix | ||
| ElephantBench | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Eleusis | Agents and tools | 9 of 9 | Nothing to fix | ||
| ELI5 Arena | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Embodied First Exam | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| Embodied Over-Refusal Benchmark | Safety and security | 9 of 9 | Nothing to fix | ||
| EmbodiedLab | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| EmbSpatialBench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| EMDB | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Emotional Attachment evaluation | Safety and security | 5 of 5 | Nothing to fix | ||
| EmpatheticDialogues | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Encord Egocentric VQA Benchmark | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Encyclopedic VQA | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| EndoVis18-VQA | Professional work | 7 of 7 | Nothing to fix | ||
| EngiWorld | Professional work | 10 of 10 | Nothing to fix | ||
| English to Icelandic Test Suite for Syntactic Divergence | Knowledge and language | 8 of 8 | Nothing to fix | ||
| EnigmaEval | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Enterprise generative evaluation (Cohere) | Professional work | 3 of 3 | Nothing to fix | ||
| Enterprise RAG evaluation (Cohere) | Knowledge and language | 3 of 3 | Nothing to fix | ||
| Enterprise-Bench L1-L2 | Agents and tools | 11 of 11 | Nothing to fix | ||
| Enterprise-Bench L1-L2 v2 | Agents and tools | 9 of 9 | Nothing to fix | ||
| EnterpriseBench: CoreCraft | Professional work | 12 of 12 | Nothing to fix | ||
| EnterpriseOps-Gym | Agents and tools | 10 of 10 | Nothing to fix | ||
| EnterpriseRAG-Bench | Agents and tools | 9 of 9 | Nothing to fix | ||
| Envisioning Signals Benchmark | Professional work | 10 of 10 | Nothing to fix | ||
| EpiBench | Safety and security | 11 of 11 | Nothing to fix | ||
| EPIC-KITCHENS VISOR Hand Object Segmentation | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| EPIC-KITCHENS VISOR Semi-Supervised Video Object Segmentation | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| EPIC-KITCHENS-100 action anticipation | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Epic-tent-O | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Epoch AI no-CoT serial reasoning steps | Safety and security | 4 of 4 | Nothing to fix | ||
| Epoch Automation Reports | Professional work | 11 of 11 | Nothing to fix | ||
| EPRSTMT | Knowledge and language | 7 of 7 | Nothing to fix | ||
| EQ-Bench 3 | Knowledge and language | 9 of 9 | Nothing to fix | ||
| EQ-Bench 4 | Knowledge and language | 10 of 10 | Nothing to fix | ||
| EQ-Bench v2 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| ERQA | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| ERRnews broadcast summarization | Knowledge and language | 8 of 8 | Nothing to fix | ||
| EsBBQ | Safety and security | 8 of 8 | Nothing to fix | ||
| ESC-50 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ESCI Shopping Queries Dataset | Knowledge and language | 7 of 7 | Nothing to fix | ||
| EstNER | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Estonian Valence Corpus | Knowledge and language | 10 of 10 | Nothing to fix | ||
| ETHICS | Safety and security | 8 of 8 | Nothing to fix | ||
| EUREKA-Bench (OpenAI) | Safety and security | 4 of 4 | Nothing to fix | ||
| EuroEval | Knowledge and language | 10 of 10 | Nothing to fix | ||
| EuroEval ERRNews | Knowledge and language | 10 of 10 | Nothing to fix | ||
| European Parliament election chatbot audit | Knowledge and language | 7 of 7 | Nothing to fix | ||
| EuroSAT | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Evalatro v2 | Agents and tools | 9 of 9 | Nothing to fix | ||
| EvalCheck-LLM | Knowledge and language | 8 of 8 | Nothing to fix | ||
| EvalCrafter | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Evaluation of Essay On-Topic Graded Comments | Knowledge and language | 8 of 8 | Nothing to fix | ||
| EvaOmni | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Event-based Segmentation Dataset | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Evidence Inference 2.0 | Professional work | 8 of 8 | Nothing to fix | ||
| EvoArena PersonaMem-Evo | Knowledge and language | 8 of 8 | Nothing to fix | ||
| EvoArena SWE-Chain-Evo | Knowledge and language | 8 of 8 | Nothing to fix | ||
| EvoCode-Bench | Coding | 9 of 9 | Nothing to fix | ||
| EvoCode-Bench v2 | Coding | 10 of 10 | Nothing to fix | ||
| EvoPolicyGym | Coding | 8 of 8 | Nothing to fix | ||
| EVVE | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| EWoK | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Exact Words | Knowledge and language | 9 of 9 | Nothing to fix | ||
| EXAMS | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Excel Modeling Benchmark | Professional work | 11 of 11 | Nothing to fix | ||
| Excel Modeling Benchmark v1.1 | Professional work | 12 of 12 | Nothing to fix | ||
| ExcelFixBench | Professional work | 9 of 9 | Nothing to fix | ||
| ExFunTube | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Expected Cyber Refusals | Safety and security | 4 of 4 | Nothing to fix | ||
| Expert Red Teaming for Violent Attack Planning | Safety and security | 4 of 4 | Nothing to fix | ||
| Expert-SWE (OpenAI) | Coding | 5 of 5 | Nothing to fix | ||
| ExpertCrowd-SWE | Coding | 11 of 11 | Nothing to fix | ||
| ExploitBench - Internal Port | Safety and security | 7 of 7 | Nothing to fix | ||
| ExploitGym | Safety and security | 12 of 12 | Nothing to fix | ||
| ExplorationBench | Math and science | 9 of 9 | Nothing to fix | ||
| Extended NYT Connections | Knowledge and language | 10 of 10 | Nothing to fix | ||
| External red team jailbreaks | Safety and security | 4 of 4 | Nothing to fix | ||
| ExtractBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| ExtractBench Pro | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| FactoryBench-100 | Professional work | 9 of 9 | Nothing to fix | ||
| FACTS Benchmark Suite | Knowledge and language | 12 of 12 | Nothing to fix | ||
| FACTS Grounding | Knowledge and language | 10 of 10 | Nothing to fix | ||
| FACTS Grounding v2 | Knowledge and language | 11 of 11 | Nothing to fix | ||
| FACTS Multimodal | Knowledge and language | 11 of 11 | Nothing to fix | ||
| FACTS Parametric | Knowledge and language | 11 of 11 | Nothing to fix | ||
| FACTS Search | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Factuality Heavy prompts | Knowledge and language | 4 of 4 | Nothing to fix | ||
| fAIntasy Football | Agents and tools | 9 of 9 | Nothing to fix | ||
| FairFace | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| FairScreen: Testing LLM Hiring Decisions | Safety and security | 7 of 7 | Nothing to fix | ||
| FakeTrueBR | Knowledge and language | 7 of 7 | Nothing to fix | ||
| False-premise questions (Anthropic) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Far-Field ASR Leaderboard | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Faroese Grammatical Correctness | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Faroese Metaphorical Explanations | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Faroese Semantic Textual Similarity | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Fashion IQ | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| FDA Information Extraction (Based) | Knowledge and language | 8 of 8 | Nothing to fix | ||
| FDE-Gym | Coding | 10 of 10 | Nothing to fix | ||
| FEABench | Coding | 7 of 7 | Nothing to fix | ||
| Fechamento BR | Professional work | 8 of 8 | Nothing to fix | ||
| Feedback Prize: Evaluating Student Writing | Professional work | 8 of 8 | Nothing to fix | ||
| Feedback Prize: Predicting Effective Arguments | Professional work | 8 of 8 | Nothing to fix | ||
| FeedbackQA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| FEM-Bench | Coding | 8 of 8 | Nothing to fix | ||
| Fermat's Last Theorem in Lean (Anthropic) | Math and science | 5 of 5 | Nothing to fix | ||
| FeruzaSpeech | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| fev-bench | Math and science | 8 of 8 | Nothing to fix | ||
| FEVER | Knowledge and language | 7 of 7 | Nothing to fix | ||
| FGVC-Aircraft | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Fiction Bench (Foreverse) | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Fiction Bench (Story Hacker) | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Filtered, adversarial sample of production prompts | Safety and security | 3 of 3 | Nothing to fix | ||
| Finance Modeling | Professional work | 4 of 4 | Nothing to fix | ||
| Finance Modeling Pro | Professional work | 4 of 4 | Nothing to fix | ||
| FinanceReasoning | Professional work | 8 of 8 | Nothing to fix | ||
| Financial Coaching Bench | Professional work | 11 of 11 | Nothing to fix | ||
| Financial PhraseBank | Professional work | 9 of 9 | Nothing to fix | ||
| Financial Retrieval (Daloopa) | Professional work | 8 of 8 | Nothing to fix | ||
| Finch (FinWorkBench) | Professional work | 10 of 10 | Nothing to fix | ||
| FinCrime Agent Benchmark | Professional work | 10 of 10 | Nothing to fix | ||
| FINESSE-Bench | Professional work | 9 of 9 | Nothing to fix | ||
| FinEvaReranking | Professional work | 8 of 8 | Nothing to fix | ||
| FinFactReranking | Professional work | 8 of 8 | Nothing to fix | ||
| FinFIRST | Professional work | 10 of 10 | Nothing to fix | ||
| FinMTEB | Professional work | 8 of 8 | Nothing to fix | ||
| FinMTEB FINAL | Professional work | 8 of 8 | Nothing to fix | ||
| Finnish Parliament ASR test | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| FinQA | Professional work | 7 of 7 | Nothing to fix | ||
| FinSBD 2021 | Professional work | 8 of 8 | Nothing to fix | ||
| FinSearchComp | Professional work | 10 of 10 | Nothing to fix | ||
| FinSearchComp Greater China | Professional work | 8 of 8 | Nothing to fix | ||
| FinStepBench CorporateValuation | Professional work | 10 of 10 | Nothing to fix | ||
| FinStepBench DeepResearch | Professional work | 10 of 10 | Nothing to fix | ||
| FinStepBench LiveSearch | Professional work | 10 of 10 | Nothing to fix | ||
| FinSTS | Professional work | 8 of 8 | Nothing to fix | ||
| FiQA2018Reranking | Professional work | 8 of 8 | Nothing to fix | ||
| Firecrawl Search Bench | Agents and tools | 10 of 10 | Nothing to fix | ||
| First Proof Second Batch | Math and science | 7 of 7 | Nothing to fix | ||
| First Proof Third Batch | Math and science | 9 of 9 | Nothing to fix | ||
| FiVE-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Flame-React-Eval | Coding | 8 of 8 | Nothing to fix | ||
| FlashDexRetarget 50-motion benchmark | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| FlavourBench Complete Core | Knowledge and language | 9 of 9 | Nothing to fix | ||
| FlavourBench Powered v43 | Knowledge and language | 9 of 9 | Nothing to fix | ||
| FlenQA | Knowledge and language | 3 of 3 | Nothing to fix | ||
| FLEURS | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| Flex-RSI | Coding | 10 of 10 | Nothing to fix | ||
| Floorplan Arena | Coding | 10 of 10 | Nothing to fix | ||
| Flores | Knowledge and language | 4 of 4 | Nothing to fix | ||
| FLORES-200 | Knowledge and language | 3 of 3 | Nothing to fix | ||
| FLORES+ | Knowledge and language | 4 of 4 | Nothing to fix | ||
| FLTEval | Math and science | 10 of 10 | Nothing to fix | ||
| FoamBench | Agents and tools | 9 of 9 | Nothing to fix | ||
| FoamBench V2 | Agents and tools | 9 of 9 | Nothing to fix | ||
| FollowIR | Knowledge and language | 9 of 9 | Nothing to fix | ||
| FoNE | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Food-101 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| FoodTruck Bench | Professional work | 10 of 10 | Nothing to fix | ||
| FORB | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ForecastBench | Math and science | 10 of 10 | Nothing to fix | ||
| Forensics CTF challenges (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| Formal Conjectures | Math and science | 7 of 7 | Nothing to fix | ||
| Formal Logic Deduction | Math and science | 7 of 7 | Nothing to fix | ||
| FORTE | Professional work | 11 of 11 | Nothing to fix | ||
| FORTE-MM | Professional work | 10 of 10 | Nothing to fix | ||
| FoSent | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Founder Bench | Professional work | 9 of 9 | Nothing to fix | ||
| FPL 2026 Agentic FPGA Backend Optimization Competition | Professional work | 8 of 8 | Nothing to fix | ||
| FQuAD | Knowledge and language | 8 of 8 | Nothing to fix | ||
| FRAMES | Knowledge and language | 8 of 8 | Nothing to fix | ||
| FreiHAND | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| FreshStack | Knowledge and language | 9 of 9 | Nothing to fix | ||
| FretBench | Knowledge and language | 8 of 8 | Nothing to fix | ||
| FrogsGame post-training task | Coding | 10 of 10 | Nothing to fix | ||
| Frontier Agentic ML & DS | Coding | 10 of 10 | Nothing to fix | ||
| Frontier Agentic STEM | Math and science | 10 of 10 | Nothing to fix | ||
| Frontier Design Benchmark | Coding | 10 of 10 | Nothing to fix | ||
| Frontier3DBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| FrontierBioML | Safety and security | 10 of 10 | Nothing to fix | ||
| FrontierCAD | Professional work | 9 of 9 | Nothing to fix | ||
| FrontierCAD Assemblies | Professional work | 9 of 9 | Nothing to fix | ||
| FrontierCAD Parts | Professional work | 9 of 9 | Nothing to fix | ||
| FrontierChallenge | Math and science | 10 of 10 | Nothing to fix | ||
| FrontierCode 1.1 | Coding | 11 of 11 | Nothing to fix | ||
| FrontierCS | Coding | 9 of 9 | Nothing to fix | ||
| FrontierCS 1.1 | Coding | 10 of 10 | Nothing to fix | ||
| FrontierCS 2 | Coding | 11 of 11 | Nothing to fix | ||
| FrontierCyber | Safety and security | 5 of 5 | Nothing to fix | ||
| FrontierEE Bench | Professional work | 11 of 11 | Nothing to fix | ||
| FrontierFinance | Professional work | 11 of 11 | Nothing to fix | ||
| FrontierMath Erdős | Math and science | 12 of 12 | Nothing to fix | ||
| FrontierMath Tier 4 | Math and science | 10 of 10 | Nothing to fix | ||
| FrontierMath Tier 4 v2 | Math and science | 11 of 11 | Nothing to fix | ||
| FrontierMath Tiers 1-3 v2 | Math and science | 11 of 11 | Nothing to fix | ||
| FrontierML | Coding | 10 of 10 | Nothing to fix | ||
| FrontierML (Synthetic Sciences) | Coding | 9 of 9 | Nothing to fix | ||
| FrontierPhysics | Math and science | 11 of 11 | Nothing to fix | ||
| FrontierRSI | Coding | 11 of 11 | Nothing to fix | ||
| FrontierScience | Math and science | 5 of 5 | Nothing to fix | ||
| FrontierScience-Olympiad | Math and science | 10 of 10 | Nothing to fix | ||
| FrontierScience-Research | Math and science | 10 of 10 | Nothing to fix | ||
| FrontierSWE | Coding | 12 of 12 | Nothing to fix | ||
| FrontierSWE v1 | Coding | 10 of 10 | Nothing to fix | ||
| FSD50K | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| FSDKaggle2018 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| FSDKaggle2019 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Full-Duplex-Bench | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| Full-Duplex-Bench v1.5 | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| Full-Duplex-Bench v2 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| FunGraph3D | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Furniture Assembly Benchmark | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| GAIA | Agents and tools | 8 of 8 | Nothing to fix | ||
| GAIK Ajokortti transcription test | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Galahad | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Galileo Agent Leaderboard | Agents and tools | 8 of 8 | Nothing to fix | ||
| Galileo LLM Hallucination Index | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Game Arena Bargaining | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Checkers | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Chess | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Clobber | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Coin Game | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Dark Hex | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Dots and Boxes | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Five in a Row | Agents and tools | 9 of 9 | Nothing to fix | ||
| Game Arena Four in a Row | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Go | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Lines of Action | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Nine Men's Morris | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Poker | Agents and tools | 9 of 9 | Nothing to fix | ||
| Game Arena Reversi | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Ultimate Tic-Tac-Toe | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Word Art | Agents and tools | 10 of 10 | Nothing to fix | ||
| Game Arena Word Association | Agents and tools | 10 of 10 | Nothing to fix | ||
| GameAdsBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| GameASG-Bench | Coding | 10 of 10 | Nothing to fix | ||
| GameDevBench | Coding | 11 of 11 | Nothing to fix | ||
| GameWorld | Agents and tools | 10 of 10 | Nothing to fix | ||
| GameXpert-Bench | Coding | 9 of 9 | Nothing to fix | ||
| GAOKAO-Bench | Knowledge and language | 8 of 8 | Nothing to fix | ||
| GaP: Graph-as-Policy | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| GaRAGe | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Garble-Bench (Voice Arena) | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| GarbleBench | Knowledge and language | 11 of 11 | Nothing to fix | ||
| GBAEval | Coding | 10 of 10 | Nothing to fix | ||
| GD-VCR | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| GDM Situational Awareness | Safety and security | 5 of 5 | Nothing to fix | ||
| GDM Stealth | Safety and security | 5 of 5 | Nothing to fix | ||
| GDP.xlsx | Professional work | 11 of 11 | Nothing to fix | ||
| GDPval | Professional work | 9 of 9 | Nothing to fix | ||
| GDPval Extended | Professional work | 11 of 11 | Nothing to fix | ||
| GDPval-AA v2.1 | Professional work | 6 of 6 | Nothing to fix | ||
| GDPVal-Diamond | Professional work | 4 of 4 | Nothing to fix | ||
| GDPval-Rubrics (MiniMax) | Professional work | 5 of 5 | Nothing to fix | ||
| GDPx | Professional work | 11 of 11 | Nothing to fix | ||
| GEBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Gecko | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| GeckoNum | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| GenAI-Arena Image Editing | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| GenAI-Arena Image Generation | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| GenAI-Arena Video Generation | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| GeneBench Pro | Safety and security | 10 of 10 | Nothing to fix | ||
| General Assistant (MiniMax in-house) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| GenEval | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| GEO-Bench-2 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| GeoBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| GeoBenchX | Agents and tools | 9 of 9 | Nothing to fix | ||
| GeoBenchX v2 | Agents and tools | 11 of 11 | Nothing to fix | ||
| GeoDE | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| GeoDrive-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| GeoGuessBench (Nutlope) | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| GeoGuesser Bench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| GeoGuesser environment (FineEnvs) | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| GeoSound | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| GerDaLIR | Professional work | 8 of 8 | Nothing to fix | ||
| GerDaLIR Small | Professional work | 9 of 9 | Nothing to fix | ||
| GerLayQA | Professional work | 9 of 9 | Nothing to fix | ||
| GerLeRB | Professional work | 9 of 9 | Nothing to fix | ||
| German fruit-story antipresuppositions | Knowledge and language | 8 of 8 | Nothing to fix | ||
| GhostInk: Honest OCR on Bill Photos | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| GIFT-Eval | Math and science | 8 of 8 | Nothing to fix | ||
| GIFT-Eval-CTX | Math and science | 9 of 9 | Nothing to fix | ||
| GigaSpeech | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| GLEE | Agents and tools | 9 of 9 | Nothing to fix | ||
| Global MMLU | Knowledge and language | 6 of 6 | Nothing to fix | ||
| Global MMLU Lite 3.0 | Knowledge and language | 10 of 10 | Nothing to fix | ||
| GlobeBench | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| GLUE | Knowledge and language | 8 of 8 | Nothing to fix | ||
| GlyphBench | Agents and tools | 8 of 8 | Nothing to fix | ||
| Gold Commodity News | Professional work | 8 of 8 | Nothing to fix | ||
| Golden3 speech summarization (Microsoft) | Vision, audio and robotics | 3 of 3 | Nothing to fix | ||
| GomokuBench | Agents and tools | 9 of 9 | Nothing to fix | ||
| Google Landmarks Dataset v2 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Gorgias AI Agent Competitive Benchmark | Professional work | 8 of 8 | Nothing to fix | ||
| GovReport | Knowledge and language | 8 of 8 | Nothing to fix | ||
| GPBam | Professional work | 9 of 9 | Nothing to fix | ||
| GPU Puzzles | Coding | 7 of 7 | Nothing to fix | ||
| GQA | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Grade Your Own Homework: Can LLMs Judge Each Other | Knowledge and language | 9 of 9 | Nothing to fix | ||
| GradeBench | Professional work | 11 of 11 | Nothing to fix | ||
| Graduate Admissions Agent Benchmark | Professional work | 9 of 9 | Nothing to fix | ||
| GrailQA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Grammarly Corpus of Discourse Coherence | Knowledge and language | 8 of 8 | Nothing to fix | ||
| GraphWalks | Knowledge and language | 5 of 5 | Nothing to fix | ||
| Gray Swan Agent Red Teaming (ART) benchmark | Safety and security | 9 of 9 | Nothing to fix | ||
| Gray Swan Bio Eval | Safety and security | 4 of 4 | Nothing to fix | ||
| Gray Swan Cyber Eval | Safety and security | 4 of 4 | Nothing to fix | ||
| Gray Swan IPI Arena benchmark | Safety and security | 10 of 10 | Nothing to fix | ||
| Gray Swan Jailbreak Arena | Safety and security | 3 of 3 | Nothing to fix | ||
| Gray Swan Shade knowledge elicitation through chat | Safety and security | 4 of 4 | Nothing to fix | ||
| Greek Legal Code | Professional work | 10 of 10 | Nothing to fix | ||
| GreekMMLU | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Grid Perception Reliability | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Grooming the Machine | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Grounding robustness (RAG prompt injection) | Safety and security | 4 of 4 | Nothing to fix | ||
| Grounding with citations, sentence unfactual rate | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Groundtruth Dynamic Benchmarking (Geology Edition) | Knowledge and language | 9 of 9 | Nothing to fix | ||
| GS-QA2 | Agents and tools | 7 of 7 | Nothing to fix | ||
| GSM-Plus | Math and science | 8 of 8 | Nothing to fix | ||
| GSM-Symbolic | Math and science | 8 of 8 | Nothing to fix | ||
| GSM1k | Math and science | 8 of 8 | Nothing to fix | ||
| GSM8K Platinum Initial 268 | Math and science | 8 of 8 | Nothing to fix | ||
| GSM8K-Platinum | Math and science | 8 of 8 | Nothing to fix | ||
| GSO-Bench | Coding | 9 of 9 | Nothing to fix | ||
| GTA | Agents and tools | 9 of 9 | Nothing to fix | ||
| GTFS Retrieval | Knowledge and language | 7 of 7 | Nothing to fix | ||
| GTO Wizard Benchmark | Agents and tools | 9 of 9 | Nothing to fix | ||
| Gulf Text-to-SQL Bench | Coding | 9 of 9 | Nothing to fix | ||
| GutBrainIE 2025 | Professional work | 7 of 7 | Nothing to fix | ||
| GutBrainIE 2026 | Professional work | 7 of 7 | Nothing to fix | ||
| Gym-Anything-RT | Agents and tools | 10 of 10 | Nothing to fix | ||
| Hack the Box challenge set (Google DeepMind) | Safety and security | 4 of 4 | Nothing to fix | ||
| Hack the Box Machines (Meta) | Safety and security | 4 of 4 | Nothing to fix | ||
| HackBench | Safety and security | 9 of 9 | Nothing to fix | ||
| HAE-RAE Bench | Knowledge and language | 4 of 4 | Nothing to fix | ||
| HAE-RAE Bench 1.0 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| HalachaBench | Professional work | 10 of 10 | Nothing to fix | ||
| Hallmarks of Cancer | Safety and security | 7 of 7 | Nothing to fix | ||
| Hallucinations on production-representative prompts | Knowledge and language | 3 of 3 | Nothing to fix | ||
| HalluScoring 2026 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| HallusionBench | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Hallym University Women's Health Nursing Exam | Professional work | 8 of 8 | Nothing to fix | ||
| HaltEval-prelim (Meta) | Coding | 3 of 3 | Nothing to fix | ||
| HaluMem | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Hanabi Agent Benchmark | Agents and tools | 10 of 10 | Nothing to fix | ||
| HANDBOOK.md | Knowledge and language | 11 of 11 | Nothing to fix | ||
| HandWorldBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| HappyWorld Embodied | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| HappyWorld Spatial Reconstruction | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Harbor-Index | Agents and tools | 10 of 10 | Nothing to fix | ||
| Hard Capability (MiniMax in-house) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Hard-negative protein binding prediction | Safety and security | 5 of 5 | Nothing to fix | ||
| Hard12 | Coding | 11 of 11 | Nothing to fix | ||
| HarmBench | Safety and security | 4 of 4 | Nothing to fix | ||
| Harmful content defect rate (Azure AI Evaluation SDK, Microsoft) | Safety and security | 3 of 3 | Nothing to fix | ||
| Harmful manipulation efficacy study | Safety and security | 4 of 4 | Nothing to fix | ||
| Harmful manipulation propensity evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| Harmful manipulation strategic planning red teaming | Safety and security | 4 of 4 | Nothing to fix | ||
| Harmfulness scores on held-out prompts (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| Harmony4D | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| harness-bench-fast | Agents and tools | 8 of 8 | Nothing to fix | ||
| HarnessEval-W | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| HarnessOpt-Bench | Coding | 9 of 9 | Nothing to fix | ||
| HarvestBench | Safety and security | 9 of 9 | Nothing to fix | ||
| Harvey LAB (Legal Agent Benchmark) | Professional work | 11 of 11 | Nothing to fix | ||
| Harvey LAB v1.1 (HLAB-Vals-1.1) | Professional work | 12 of 12 | Nothing to fix | ||
| Harvey LAB-AA v1.0 | Professional work | 4 of 4 | Nothing to fix | ||
| Harvey LAB-AA v1.1 | Professional work | 6 of 6 | Nothing to fix | ||
| Hate & Harassment Safety | Safety and security | 4 of 4 | Nothing to fix | ||
| Hateful Memes Challenge | Safety and security | 8 of 8 | Nothing to fix | ||
| HC3Reranking | Professional work | 8 of 8 | Nothing to fix | ||
| HEAD-QA | Professional work | 7 of 7 | Nothing to fix | ||
| Health dynamic conversation eval | Professional work | 4 of 4 | Nothing to fix | ||
| Health language and locale eval | Professional work | 4 of 4 | Nothing to fix | ||
| Health static understanding eval | Professional work | 4 of 4 | Nothing to fix | ||
| HealthAdvice | Professional work | 7 of 7 | Nothing to fix | ||
| HealthAgentBench | Professional work | 10 of 10 | Nothing to fix | ||
| HealthAgentBench V2 | Professional work | 11 of 11 | Nothing to fix | ||
| HealthBench | Professional work | 11 of 11 | Nothing to fix | ||
| HealthBench Consensus | Professional work | 10 of 10 | Nothing to fix | ||
| HealthBench Hard | Professional work | 11 of 11 | Nothing to fix | ||
| HealthBench-Psych v1 | Professional work | 9 of 9 | Nothing to fix | ||
| HealthBench-Psych v2 | Professional work | 10 of 10 | Nothing to fix | ||
| HealthBench-Psych-Hard | Professional work | 10 of 10 | Nothing to fix | ||
| HealthQA-BR | Professional work | 8 of 8 | Nothing to fix | ||
| HearingBench | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| Hebrew Diacritization Test Corpora | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Hedge-Bench | Professional work | 9 of 9 | Nothing to fix | ||
| HEIM | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Helpfulness violations (from ART) | Safety and security | 3 of 3 | Nothing to fix | ||
| Hemingway-bench | Knowledge and language | 9 of 9 | Nothing to fix | ||
| HEPToolBench 1.2 | Math and science | 10 of 10 | Nothing to fix | ||
| HeQ | Knowledge and language | 8 of 8 | Nothing to fix | ||
| HexMachina | Agents and tools | 7 of 7 | Nothing to fix | ||
| Hi4D | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Hidden Voices Audio Difficulty | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| HiddenMath | Math and science | 6 of 6 | Nothing to fix | ||
| HiddenMath-Hard | Math and science | 4 of 4 | Nothing to fix | ||
| HiDrive | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| HierText | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| High Stakes factuality prompts | Knowledge and language | 4 of 4 | Nothing to fix | ||
| High-risk health scenarios eval | Professional work | 4 of 4 | Nothing to fix | ||
| Higher-difficulty benign request evaluations (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| Higher-difficulty harmful request evaluations (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| HIPE-OCRepair 2026 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| HiPhO | Math and science | 9 of 9 | Nothing to fix | ||
| Historical Jailbreak Regression | Safety and security | 4 of 4 | Nothing to fix | ||
| HLE Calibration | Knowledge and language | 9 of 9 | Nothing to fix | ||
| HLE Extended | Knowledge and language | 10 of 10 | Nothing to fix | ||
| HLE-Verified | Knowledge and language | 10 of 10 | Nothing to fix | ||
| HLS-Eval | Professional work | 7 of 7 | Nothing to fix | ||
| HM3DSem-Walks | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| HMMT 2025 | Math and science | 9 of 9 | Nothing to fix | ||
| HMMT 2026 | Math and science | 10 of 10 | Nothing to fix | ||
| HMMT Nov 2025 | Math and science | 10 of 10 | Nothing to fix | ||
| HMU-Bench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Hold or Fold: AI Under Pushback | Safety and security | 9 of 9 | Nothing to fix | ||
| Holdings-IndividualTaxBench | Professional work | 11 of 11 | Nothing to fix | ||
| HoloAssist Mistake Detection | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Honesty human feedback win rate (Anthropic) | Knowledge and language | 5 of 5 | Nothing to fix | ||
| Honeypot exploitation monitorability | Safety and security | 4 of 4 | Nothing to fix | ||
| Honor of Kings Bug AI Root Cause Analysis | Professional work | 9 of 9 | Nothing to fix | ||
| HorizonMath | Math and science | 10 of 10 | Nothing to fix | ||
| HotpotQA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| HourVideo | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| How2QA | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| HR-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| HRM8K | Math and science | 9 of 9 | Nothing to fix | ||
| HRM8K-KSM | Math and science | 9 of 9 | Nothing to fix | ||
| HuggingFace Agent Benchmark | Agents and tools | 3 of 3 | Nothing to fix | ||
| HUMAINE | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Human ChatGPT Comparison Corpus | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Human or AI? (Skills IL) | Knowledge and language | 11 of 11 | Nothing to fix | ||
| Human proximity (Google DeepMind robotics) | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Human Sourced Jailbreaks | Safety and security | 4 of 4 | Nothing to fix | ||
| Human SxS image editing Elo (Google) | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| Human SxS image generation Elo (Google) | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| Human Verification Gap | Safety and security | 10 of 10 | Nothing to fix | ||
| HumanCLAW-Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| HumanEval+ | Coding | 9 of 9 | Nothing to fix | ||
| Humanity's Sixth Sense | Vision, audio and robotics | 12 of 12 | Nothing to fix | ||
| HumanoidBench | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| HumanoidToolBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Hume AI Overall Quality Index | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| Hume AI Voice Design Benchmark | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| HumMusQA | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Hungarian Riddles Benchmark | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Hunyuan-MT Challenge Testset (Tencent) | Knowledge and language | 3 of 3 | Nothing to fix | ||
| HW-Bench | Professional work | 11 of 11 | Nothing to fix | ||
| HWE Bench V1 (RISC-V CPU design) | Professional work | 9 of 9 | Nothing to fix | ||
| HWE Bench V2 (RISC-V CPU design) | Professional work | 10 of 10 | Nothing to fix | ||
| HWE-bench CAE | Professional work | 10 of 10 | Nothing to fix | ||
| HWE-Bench v1.0 (RTL bug repair) | Professional work | 10 of 10 | Nothing to fix | ||
| HWE-Bench v1.1 (RTL bug repair) | Professional work | 10 of 10 | Nothing to fix | ||
| Hy-Backend | Coding | 4 of 4 | Nothing to fix | ||
| Hy-Backend 2.0 | Coding | 5 of 5 | Nothing to fix | ||
| Hy-BrowseComp-Pro2 | Agents and tools | 5 of 5 | Nothing to fix | ||
| Hy-CompanyBench | Coding | 5 of 5 | Nothing to fix | ||
| Hy-CompanyBench V2 | Coding | 5 of 5 | Nothing to fix | ||
| Hy-Euler pro | Math and science | 5 of 5 | Nothing to fix | ||
| Hy-FinAgentBench | Professional work | 5 of 5 | Nothing to fix | ||
| Hy-FinModelBench | Professional work | 5 of 5 | Nothing to fix | ||
| Hy-FinmodelBench v2 | Professional work | 5 of 5 | Nothing to fix | ||
| Hy-LifeSearch | Agents and tools | 5 of 5 | Nothing to fix | ||
| Hy-Math | Math and science | 5 of 5 | Nothing to fix | ||
| Hy-SkillsWorld | Agents and tools | 5 of 5 | Nothing to fix | ||
| Hy-SWE Max | Coding | 5 of 5 | Nothing to fix | ||
| Hy-Vibe Bench | Coding | 4 of 4 | Nothing to fix | ||
| Hype Trap | Math and science | 9 of 9 | Nothing to fix | ||
| HyperTask | Coding | 4 of 4 | Nothing to fix | ||
| ICASSP 2026 Automatic Song Aesthetics Evaluation Challenge | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ICDAR 2024 MapText Competition | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Icelandic WikiQA | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Icelandic WinoGrande | Knowledge and language | 9 of 9 | Nothing to fix | ||
| ICLayout-Bench | Professional work | 10 of 10 | Nothing to fix | ||
| IDP Core Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| IF-Badcase | Knowledge and language | 4 of 4 | Nothing to fix | ||
| IFBench | Knowledge and language | 9 of 9 | Nothing to fix | ||
| IFC-Bench v2 | Professional work | 11 of 11 | Nothing to fix | ||
| IFEval | Knowledge and language | 10 of 10 | Nothing to fix | ||
| IFEval (Mistral internal version) | Knowledge and language | 3 of 3 | Nothing to fix | ||
| IFStruct | Knowledge and language | 9 of 9 | Nothing to fix | ||
| IIT Madras Hindi ASR Challenge | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| IIT Madras Hindi-Tamil-English ASR Challenge | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| IlùBench | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Image generation refusals | Safety and security | 3 of 3 | Nothing to fix | ||
| Image input safety evaluations | Safety and security | 4 of 4 | Nothing to fix | ||
| Image Matching Challenge | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Image2FloorPlan | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| ImageCoDe | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ImageJevBench v0.1.5 | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| ImageJevBench v0.3.0 | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| ImageNet ReaL | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ImageNet-1K | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ImageNet-A | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ImageNet-C | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| ImageNet-R | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ImageNet-Sketch | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ImageNet-UA | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ImageNetV2 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| IMC 2025 | Math and science | 7 of 7 | Nothing to fix | ||
| ImgMT | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Immigration Bench | Professional work | 11 of 11 | Nothing to fix | ||
| IMO 2025 | Math and science | 10 of 10 | Nothing to fix | ||
| IMO 2026 | Math and science | 4 of 4 | Nothing to fix | ||
| IMO-AnswerBench | Math and science | 9 of 9 | Nothing to fix | ||
| IMO-ProofBench | Math and science | 9 of 9 | Nothing to fix | ||
| Impossible tasks (Anthropic) | Safety and security | 5 of 5 | Nothing to fix | ||
| Impossible Tasks Evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| In-context data exfiltration, visual browser | Safety and security | 4 of 4 | Nothing to fix | ||
| IN22-Gen | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Inappropriate Image Prompts | Safety and security | 7 of 7 | Nothing to fix | ||
| INAS hypothesis classification | Math and science | 8 of 8 | Nothing to fix | ||
| iNaturalist 2017 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| iNaturalist 2021 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| iNaturalist Sounds | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Incident Arena | Agents and tools | 11 of 11 | Nothing to fix | ||
| INCLUDE | Knowledge and language | 9 of 9 | Nothing to fix | ||
| IncompeBench | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| IndicLID benchmark | Knowledge and language | 9 of 9 | Nothing to fix | ||
| IndicQA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| IndicXNLI | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Indirect prompt injection, Actor Critic attack | Safety and security | 3 of 3 | Nothing to fix | ||
| Indirect prompt injection, Beam Search attack | Safety and security | 3 of 3 | Nothing to fix | ||
| Indirect prompt injection, TAP attack | Safety and security | 3 of 3 | Nothing to fix | ||
| IndiVibe | Knowledge and language | 3 of 3 | Nothing to fix | ||
| IndoNLI | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Inductive Arithmetic Pattern Reasoning | Math and science | 10 of 10 | Nothing to fix | ||
| InferenceBench | Coding | 10 of 10 | Nothing to fix | ||
| Infinite Generations (Mistral internal) | Knowledge and language | 3 of 3 | Nothing to fix | ||
| Infographic factuality (Google) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| InfographicVQA | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Information Extraction (ByteDance Seed in-house) | Professional work | 4 of 4 | Nothing to fix | ||
| Information Processing (ByteDance Seed in-house) | Professional work | 4 of 4 | Nothing to fix | ||
| Infrathon | Coding | 10 of 10 | Nothing to fix | ||
| INQUIRE-Fullrank | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| INQUISITIVE | Knowledge and language | 8 of 8 | Nothing to fix | ||
| InSAR Regional Snow Mapping (Axis Spatial) | Math and science | 9 of 9 | Nothing to fix | ||
| Instruction Following (MiniMax in-house) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Instruction hierarchy attacks | Safety and security | 5 of 5 | Nothing to fix | ||
| InstructIR | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Instrument reading (Google DeepMind) | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Instrumental convergence evals | Safety and security | 4 of 4 | Nothing to fix | ||
| Instrumentation Universal Knowledge v1.3 | Professional work | 8 of 8 | Nothing to fix | ||
| Instrumentation Universal Knowledge v1.5 | Professional work | 8 of 8 | Nothing to fix | ||
| Instrumentation Universal Knowledge v1.6 refresh | Professional work | 10 of 10 | Nothing to fix | ||
| Insurance Agent Benchmark | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| InsureBench Decision Models (Broader Comparison) | Professional work | 8 of 8 | Nothing to fix | ||
| InsureBench Multi-functional LLMs | Professional work | 9 of 9 | Nothing to fix | ||
| Intention Recognition (ByteDance Seed in-house) | Professional work | 4 of 4 | Nothing to fix | ||
| InteractiveBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| InterCode-CTF | Safety and security | 8 of 8 | Nothing to fix | ||
| Internal adversarially-designed factuality evaluations | Knowledge and language | 5 of 5 | Nothing to fix | ||
| Internal agentic coding evaluation (Anthropic) | Coding | 4 of 4 | Nothing to fix | ||
| Internal AI research evaluation suite 2 | Coding | 4 of 4 | Nothing to fix | ||
| Internal codebase monitor evasion eval | Safety and security | 4 of 4 | Nothing to fix | ||
| Internal codebase sabotage propensity (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| Internal computer use safety benchmark (OpenAI) | Agents and tools | 4 of 4 | Nothing to fix | ||
| Internal model use survey (Anthropic) | Coding | 5 of 5 | Nothing to fix | ||
| Internal Research Debugging Evaluation | Coding | 7 of 7 | Nothing to fix | ||
| Internal vulnerability research | Safety and security | 5 of 5 | Nothing to fix | ||
| IntervalQA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Into the Unknown | Math and science | 10 of 10 | Nothing to fix | ||
| IntPhys 2 | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| IntPhys2-Timing | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| INVALSI Italian LLM Benchmark | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Invented O'Clock | Knowledge and language | 8 of 8 | Nothing to fix | ||
| IOI (Vals) | Coding | 9 of 9 | Nothing to fix | ||
| IOL-AI 2026 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| IPI Bench | Safety and security | 10 of 10 | Nothing to fix | ||
| Iris Safety Evaluations | Safety and security | 4 of 4 | Nothing to fix | ||
| Irrelevant instructions, text-based web browser | Safety and security | 3 of 3 | Nothing to fix | ||
| Irrelevant instructions, visual browser | Safety and security | 3 of 3 | Nothing to fix | ||
| iSAID | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Ishigaki-IDS-Bench | Professional work | 11 of 11 | Nothing to fix | ||
| IslamicEval 2025 Quotation Correction | Knowledge and language | 7 of 7 | Nothing to fix | ||
| IslamicEval 2025 Quotation Span Identification | Knowledge and language | 8 of 8 | Nothing to fix | ||
| IslamicEval 2025 Quotation Validation | Knowledge and language | 7 of 7 | Nothing to fix | ||
| IslamicEval 2025 Quran and Hadith Passage Retrieval | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Italian-Administrative-CaseHold | Professional work | 10 of 10 | Nothing to fix | ||
| ITSMBench | Agents and tools | 9 of 9 | Nothing to fix | ||
| Ivo-micro1 ContractBench | Professional work | 11 of 11 | Nothing to fix | ||
| Jailbreak Augmented Examples | Safety and security | 4 of 4 | Nothing to fix | ||
| JailTrickBench | Safety and security | 7 of 7 | Nothing to fix | ||
| Jam-ALT | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| JAMA Clinical Challenge (JesseLiu) | Professional work | 8 of 8 | Nothing to fix | ||
| JamendoLyrics | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| JamendoLyrics MultiLang | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Jamp | Knowledge and language | 8 of 8 | Nothing to fix | ||
| JaNLI | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Japanese National Dental Examination 2024 | Professional work | 8 of 8 | Nothing to fix | ||
| Japanese National Examination for Occupational Therapists 2025 | Professional work | 7 of 7 | Nothing to fix | ||
| JASMINE | Professional work | 7 of 7 | Nothing to fix | ||
| JBLiMP | Knowledge and language | 7 of 7 | Nothing to fix | ||
| JEEM | Knowledge and language | 7 of 7 | Nothing to fix | ||
| JEF-Bench | Safety and security | 4 of 4 | Nothing to fix | ||
| Jester gesture recognition | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| JevBench | Other | 9 of 9 | Nothing to fix | ||
| JFBench | Knowledge and language | 10 of 10 | Nothing to fix | ||
| JFinBench | Professional work | 9 of 9 | Nothing to fix | ||
| JGLUE | Knowledge and language | 8 of 8 | Nothing to fix | ||
| JMMLU | Knowledge and language | 4 of 4 | Nothing to fix | ||
| JobBench | Professional work | 11 of 11 | Nothing to fix | ||
| JobBERT Job Title Normalization | Professional work | 8 of 8 | Nothing to fix | ||
| JointAVBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Joke Arena | Knowledge and language | 10 of 10 | Nothing to fix | ||
| JSICK | Knowledge and language | 9 of 9 | Nothing to fix | ||
| JSONSchemaBench | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Judge Bait | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Judgemark v2.1 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| JudgerBench | Other | 8 of 8 | Nothing to fix | ||
| JudgmentBench | Professional work | 9 of 9 | Nothing to fix | ||
| JusticeAxis | Professional work | 10 of 10 | Nothing to fix | ||
| K-Bench | Safety and security | 9 of 9 | Nothing to fix | ||
| K-MetBench | Math and science | 9 of 9 | Nothing to fix | ||
| K-MHaS | Safety and security | 8 of 8 | Nothing to fix | ||
| K-Sort Arena | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| KAET (Kimi Autonomous Execution Tasks) | Agents and tools | 5 of 5 | Nothing to fix | ||
| Kaggle ASCII Arena | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Kaggle Explode Arena | Coding | 10 of 10 | Nothing to fix | ||
| Kaggle Game Arena | Agents and tools | 10 of 10 | Nothing to fix | ||
| Kaggle Landmark Arena | Coding | 10 of 10 | Nothing to fix | ||
| Kaggle Pelican Arena | Coding | 10 of 10 | Nothing to fix | ||
| Kaggle Sort Viz Arena | Coding | 10 of 10 | Nothing to fix | ||
| Kaggle Standardized Agent Exams | Safety and security | 8 of 8 | Nothing to fix | ||
| Kaggle SVG Arena | Coding | 10 of 10 | Nothing to fix | ||
| KAT Claw Bench | Agents and tools | 4 of 4 | Nothing to fix | ||
| KAT Code Bench | Coding | 5 of 5 | Nothing to fix | ||
| Kazakh Speech Corpus | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| KBL | Professional work | 5 of 5 | Nothing to fix | ||
| Keep the number you wrote | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Kernel Bench L3 | Coding | 4 of 4 | Nothing to fix | ||
| KernelBench (kernelbench.com) | Coding | 9 of 9 | Nothing to fix | ||
| KernelGen 1P | Coding | 7 of 7 | Nothing to fix | ||
| KernelQuest | Coding | 9 of 9 | Nothing to fix | ||
| Kernels task (Anthropic) | Coding | 4 of 4 | Nothing to fix | ||
| Key insight rediscovery on an unpublished ML task (external testing) | Coding | 4 of 4 | Nothing to fix | ||
| KGC-Safety (LG) | Safety and security | 4 of 4 | Nothing to fix | ||
| KgCLUE | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Kiji PII Detection | Knowledge and language | 8 of 8 | Nothing to fix | ||
| KiloGram | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Kimi Claw 24/7 Bench | Professional work | 10 of 10 | Nothing to fix | ||
| Kimi Code Bench v2 | Coding | 5 of 5 | Nothing to fix | ||
| Kimi exploit development suite | Safety and security | 5 of 5 | Nothing to fix | ||
| Kimi Hallu Bench | Knowledge and language | 11 of 11 | Nothing to fix | ||
| Kimi Webdev Bench | Coding | 5 of 5 | Nothing to fix | ||
| KINA | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Kinayat | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Kinetics-400 | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Kinetics-600 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| KITTI depth prediction | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| KlinikeBench | Professional work | 10 of 10 | Nothing to fix | ||
| KMMLU-Pro | Knowledge and language | 11 of 11 | Nothing to fix | ||
| Knowledge Q&A (MiniMax in-house) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Ko-AIME 2024 (Upstage) | Math and science | 4 of 4 | Nothing to fix | ||
| Ko-GDPval | Professional work | 9 of 9 | Nothing to fix | ||
| Ko-IFEval (Upstage) | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Ko-LongBench (LG) | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Koala Test Set | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Kompetencer | Professional work | 8 of 8 | Nothing to fix | ||
| KoMT-Bench | Knowledge and language | 8 of 8 | Nothing to fix | ||
| KORA | Safety and security | 10 of 10 | Nothing to fix | ||
| KORA Apps | Safety and security | 8 of 8 | Nothing to fix | ||
| Korean Contest Prize Reading | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Korean Sarcasm Dataset | Knowledge and language | 10 of 10 | Nothing to fix | ||
| KorMedMCQA | Professional work | 9 of 9 | Nothing to fix | ||
| KorMedMCQA-Mixed | Professional work | 8 of 8 | Nothing to fix | ||
| KorMedMCQA-V | Professional work | 8 of 8 | Nothing to fix | ||
| KorNAT | Knowledge and language | 10 of 10 | Nothing to fix | ||
| KRETA | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| KRYT Benchmark | Agents and tools | 11 of 11 | Nothing to fix | ||
| KSAA 2026 Automatic Diacritization of Speech Dictation | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| KUMO | Agents and tools | 8 of 8 | Nothing to fix | ||
| KuralHub | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| KWV Bench | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| LAB-Bench | Safety and security | 10 of 10 | Nothing to fix | ||
| LAB-Bench CloningScenarios | Safety and security | 8 of 8 | Nothing to fix | ||
| LAB-Bench FigQA | Safety and security | 10 of 10 | Nothing to fix | ||
| LAB-Bench LitQA2 | Safety and security | 9 of 9 | Nothing to fix | ||
| LAB-Bench ProtocolQA | Safety and security | 9 of 9 | Nothing to fix | ||
| LAB-Bench SeqQA | Safety and security | 5 of 5 | Nothing to fix | ||
| LABBench2 | Math and science | 9 of 9 | Nothing to fix | ||
| LABBench2 LitQA3 | Safety and security | 8 of 8 | Nothing to fix | ||
| LABBench2 PatentQA | Math and science | 8 of 8 | Nothing to fix | ||
| LABBench2 SourceQuality | Math and science | 8 of 8 | Nothing to fix | ||
| LABBench2 TableQA2 (image) | Math and science | 8 of 8 | Nothing to fix | ||
| LABBench2 TableQA2 (PDF) | Math and science | 8 of 8 | Nothing to fix | ||
| LABBench2 TrialQA | Math and science | 8 of 8 | Nothing to fix | ||
| Labelbox Complex Reasoning | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Labelbox Speech Generation v2 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Labelbox Video Generation v2 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| LADDER | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Lahjoita puhetta ASR test | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Lakera B3 (Backbone Breaker Benchmark) | Safety and security | 9 of 9 | Nothing to fix | ||
| Lakh MIDI MASD Style Classification | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Lakh MIDI Top-MAGD Genre Classification | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Lakh, Crore and Carpet Area | Math and science | 8 of 8 | Nothing to fix | ||
| LAMBADA | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Landing Page Arena | Coding | 10 of 10 | Nothing to fix | ||
| Language Barrier Benchmark | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Language Confusion Benchmark | Knowledge and language | 3 of 3 | Nothing to fix | ||
| Language-Table | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Large Language Model Football League | Agents and tools | 8 of 8 | Nothing to fix | ||
| Large Movie Review Dataset | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Large-Vocabulary Video Instance Segmentation | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Last Secure Code Benchmark | Safety and security | 10 of 10 | Nothing to fix | ||
| LatchBio Capabilities V0 | Safety and security | 11 of 11 | Nothing to fix | ||
| Latin Synonym Selection Benchmark | Knowledge and language | 7 of 7 | Nothing to fix | ||
| LayMaker MTSamples Simplification | Professional work | 8 of 8 | Nothing to fix | ||
| Lazy investigation (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| Lazy investigation on misleading codebases (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| LBPP (Less Basic Python Problems) | Coding | 10 of 10 | Nothing to fix | ||
| LeanDojo Benchmark | Math and science | 7 of 7 | Nothing to fix | ||
| LeanLean | Math and science | 11 of 11 | Nothing to fix | ||
| Lech Mazur Creative Writing | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Leetcode (easy) | Coding | 4 of 4 | Nothing to fix | ||
| Leetcode (hard) | Coding | 4 of 4 | Nothing to fix | ||
| Leetcode (medium) | Coding | 4 of 4 | Nothing to fix | ||
| Legacy-Bench | Coding | 10 of 10 | Nothing to fix | ||
| LegacyShop | Coding | 10 of 10 | Nothing to fix | ||
| Legal Consent Sweep | Professional work | 9 of 9 | Nothing to fix | ||
| Legal Information Retrieval Benchmark | Professional work | 7 of 7 | Nothing to fix | ||
| Legal Opinion Sentiment Classification | Professional work | 7 of 7 | Nothing to fix | ||
| Legal Research Bench | Professional work | 12 of 12 | Nothing to fix | ||
| LegalBench | Professional work | 9 of 9 | Nothing to fix | ||
| LegalBench.PT | Professional work | 8 of 8 | Nothing to fix | ||
| LegalQuAD | Professional work | 9 of 9 | Nothing to fix | ||
| LegalReasonBench | Professional work | 11 of 11 | Nothing to fix | ||
| Legora BAR | Professional work | 10 of 10 | Nothing to fix | ||
| LengthCtrl (Qwen) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| LevelBench-100 v4.1 | Professional work | 11 of 11 | Nothing to fix | ||
| LevelBench-100 v5.0 | Professional work | 10 of 10 | Nothing to fix | ||
| LEVIR-CC | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| LexGLUE | Professional work | 8 of 8 | Nothing to fix | ||
| LexPilot | Professional work | 10 of 10 | Nothing to fix | ||
| LEXTREME | Professional work | 8 of 8 | Nothing to fix | ||
| LGEB | Knowledge and language | 7 of 7 | Nothing to fix | ||
| LIBERO | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| LibraryDesignBench | Coding | 10 of 10 | Nothing to fix | ||
| LightGenBench | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| LIMIT | Knowledge and language | 8 of 8 | Nothing to fix | ||
| LIMMITS 2024 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| LIMMITS 2025 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| LineEX | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| LingoQA | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Linguini | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Linguistic Knowledge Benchmark (Fab AI) | Knowledge and language | 9 of 9 | Nothing to fix | ||
| LinuxArena | Safety and security | 10 of 10 | Nothing to fix | ||
| LisanBench | Agents and tools | 10 of 10 | Nothing to fix | ||
| LitQA v0 | Math and science | 9 of 9 | Nothing to fix | ||
| LitSearch | Knowledge and language | 8 of 8 | Nothing to fix | ||
| LiveBench | Other | 10 of 10 | Nothing to fix | ||
| LiveClawBench | Agents and tools | 10 of 10 | Nothing to fix | ||
| LiveCodeBench | Coding | 9 of 9 | Nothing to fix | ||
| LiveCodeBench Pro | Coding | 9 of 9 | Nothing to fix | ||
| LiveCodeBench v5 | Coding | 10 of 10 | Nothing to fix | ||
| LiveCodeBench v6 | Coding | 10 of 10 | Nothing to fix | ||
| LiveMathematicianBench | Math and science | 10 of 10 | Nothing to fix | ||
| LiveNewsBench | Knowledge and language | 9 of 9 | Nothing to fix | ||
| LiveSQLBench | Coding | 11 of 11 | Nothing to fix | ||
| LlamaTouch | Agents and tools | 7 of 7 | Nothing to fix | ||
| LLM 3D Worlds Arena | Coding | 10 of 10 | Nothing to fix | ||
| LLM Agronomy Benchmark (Farmers Business Network edition) | Professional work | 10 of 10 | Nothing to fix | ||
| LLM Agronomy Benchmark (June 2024) | Professional work | 8 of 8 | Nothing to fix | ||
| LLM Agronomy Benchmark (original) | Professional work | 9 of 9 | Nothing to fix | ||
| LLM Agronomy Benchmark Image Identification (four choices) | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| LLM Benchmark: Danske Idiomer | Knowledge and language | 8 of 8 | Nothing to fix | ||
| LLM Chess | Agents and tools | 10 of 10 | Nothing to fix | ||
| LLM Chess Arena (Louis Guichard) | Agents and tools | 9 of 9 | Nothing to fix | ||
| LLM Debate Benchmark | Knowledge and language | 10 of 10 | Nothing to fix | ||
| LLM Fantasy Football Arena Classic | Agents and tools | 9 of 9 | Nothing to fix | ||
| LLM Fantasy Football Arena Guillotine | Agents and tools | 9 of 9 | Nothing to fix | ||
| LLM Physics Plotting Benchmark | Math and science | 9 of 9 | Nothing to fix | ||
| LLM Position Bias Benchmark | Safety and security | 9 of 9 | Nothing to fix | ||
| LLM SoccerArena World Cup 2026 | Math and science | 9 of 9 | Nothing to fix | ||
| LLM Sycophancy Benchmark | Safety and security | 9 of 9 | Nothing to fix | ||
| LLM training task (Anthropic) | Coding | 4 of 4 | Nothing to fix | ||
| LLM-CAE-Benchmark | Coding | 8 of 8 | Nothing to fix | ||
| LLM-IOAA | Math and science | 8 of 8 | Nothing to fix | ||
| LM-SynEval | Knowledge and language | 7 of 7 | Nothing to fix | ||
| LMArcade | Coding | 9 of 9 | Nothing to fix | ||
| LMArena Image-to-Video Arena | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| LMArena Search Arena | Agents and tools | 11 of 11 | Nothing to fix | ||
| LMArena Text-to-Video Arena | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| LMBuild | Professional work | 10 of 10 | Nothing to fix | ||
| LMCA | Other | 11 of 11 | Nothing to fix | ||
| LMSYS Chatbot Arena Human Preference Predictions | Knowledge and language | 8 of 8 | Nothing to fix | ||
| LOCA-Bench | Knowledge and language | 10 of 10 | Nothing to fix | ||
| LoCoMo | Knowledge and language | 8 of 8 | Nothing to fix | ||
| LOFT | Knowledge and language | 5 of 5 | Nothing to fix | ||
| LogicVal | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| LogicVista | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| Long Context (MiniMax in-house) | Knowledge and language | 3 of 3 | Nothing to fix | ||
| Long Context Navigator | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Long Range Arena | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Long-context grounding, sentence unfactual rate | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Long-form Biological Risk Questions | Safety and security | 3 of 3 | Nothing to fix | ||
| Long-form virology tasks | Safety and security | 5 of 5 | Nothing to fix | ||
| Long-Horizon Freight AuditBench Challenge | Professional work | 10 of 10 | Nothing to fix | ||
| Long-Horizon Freight AuditBench Production | Professional work | 10 of 10 | Nothing to fix | ||
| Long-Horizon Medical Agent Benchmark, Veterinary Clinic Ops v2 | Professional work | 9 of 9 | Nothing to fix | ||
| Long-Horizon-Terminal-Bench (LHTB) | Agents and tools | 10 of 10 | Nothing to fix | ||
| Long-Horizon-Terminal-Bench 2.0 | Agents and tools | 10 of 10 | Nothing to fix | ||
| Long-term Video Object Segmentation v2 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| LongBench | Knowledge and language | 4 of 4 | Nothing to fix | ||
| LongBench Chinese | Knowledge and language | 8 of 8 | Nothing to fix | ||
| LongBench v2 | Knowledge and language | 11 of 11 | Nothing to fix | ||
| LongCat safety evaluation (Meituan) | Safety and security | 3 of 3 | Nothing to fix | ||
| LongFact | Knowledge and language | 5 of 5 | Nothing to fix | ||
| Longform Creative Writing (EQ-Bench) | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Longform Writing | Knowledge and language | 4 of 4 | Nothing to fix | ||
| LongMemEval | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Longtail-SWE | Coding | 10 of 10 | Nothing to fix | ||
| LongVideoBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| LoopArena | Coding | 10 of 10 | Nothing to fix | ||
| Loss Run Extraction (Bevaya) | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| LoTTE | Knowledge and language | 8 of 8 | Nothing to fix | ||
| LoveDA | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| LOVEU-TGVE-2023 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| LSMDC | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| Ludobench Backgammon | Agents and tools | 9 of 9 | Nothing to fix | ||
| Ludobench Hokm | Agents and tools | 9 of 9 | Nothing to fix | ||
| LVBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| LVIS | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| LVIS-MG | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| LVOmniBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Lying about identity (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| m-ArenaHard | Knowledge and language | 8 of 8 | Nothing to fix | ||
| M-IFEval | Knowledge and language | 8 of 8 | Nothing to fix | ||
| m-tau-bench (Cohere) | Agents and tools | 4 of 4 | Nothing to fix | ||
| m-WildVision | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| M³AV Speech Recognition | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| M5HisDoc | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MACHIAVELLI | Safety and security | 7 of 7 | Nothing to fix | ||
| MacroscopeBench | Coding | 11 of 11 | Nothing to fix | ||
| MADASR 2.0 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MADASR 2023 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| MADQA | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| MAEB | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MAESTRO Piano Transcription | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MAI human side-by-side evaluation | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Make Data Count: Finding Data References | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Malicious computer use (Anthropic) | Safety and security | 6 of 6 | Nothing to fix | ||
| Malicious use of Claude Code (Anthropic) | Safety and security | 6 of 6 | Nothing to fix | ||
| Malware Refusals (Golden Set) | Safety and security | 3 of 3 | Nothing to fix | ||
| Malware Refusals (Synthetic) | Safety and security | 3 of 3 | Nothing to fix | ||
| MammalNet | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Mana RP Bench | Knowledge and language | 10 of 10 | Nothing to fix | ||
| MANDOS Run 003 | Professional work | 8 of 8 | Nothing to fix | ||
| MANDOS Session 2 | Professional work | 8 of 8 | Nothing to fix | ||
| ManiSkill3 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| ManyIH-Bench | Knowledge and language | 10 of 10 | Nothing to fix | ||
| MAPA EUR-Lex | Professional work | 7 of 7 | Nothing to fix | ||
| MapAnything single-view calibration | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| MarsRetrieval | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| MasakhaNER | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MasakhaNER 2.0 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MASK | Safety and security | 9 of 9 | Nothing to fix | ||
| MASSIVE | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Massive Visualized IR Benchmark | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MATH | Math and science | 9 of 9 | Nothing to fix | ||
| Math Kangaroo 2025 (MathArena) | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Math Side-tasking Monitor Evasion | Safety and security | 4 of 4 | Nothing to fix | ||
| MATH-500 | Math and science | 10 of 10 | Nothing to fix | ||
| MATH-Vision (MathVision) | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Math24o | Math and science | 8 of 8 | Nothing to fix | ||
| MathArena (Apex, IMO, USAMO) | Math and science | 10 of 10 | Nothing to fix | ||
| MathArena Apex Shortlist | Math and science | 10 of 10 | Nothing to fix | ||
| MathTutorBench | Math and science | 9 of 9 | Nothing to fix | ||
| MathVerse | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| MathVista | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MatSci-NLP | Math and science | 8 of 8 | Nothing to fix | ||
| MAUD | Professional work | 8 of 8 | Nothing to fix | ||
| MAWARITH | Professional work | 8 of 8 | Nothing to fix | ||
| MAXIFE | Knowledge and language | 5 of 5 | Nothing to fix | ||
| MaxxBench | Coding | 11 of 11 | Nothing to fix | ||
| Maze Bench | Agents and tools | 11 of 11 | Nothing to fix | ||
| MBABench | Professional work | 8 of 8 | Nothing to fix | ||
| MBABench 2.0 | Professional work | 10 of 10 | Nothing to fix | ||
| MBench | Other | 8 of 8 | Nothing to fix | ||
| MBPP | Coding | 9 of 9 | Nothing to fix | ||
| MBXP | Coding | 4 of 4 | Nothing to fix | ||
| MC-TACO | Knowledge and language | 7 of 7 | Nothing to fix | ||
| MCP Atlas | Agents and tools | 10 of 10 | Nothing to fix | ||
| MCP prompt injection evaluation (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| MCPMark | Agents and tools | 10 of 10 | Nothing to fix | ||
| MCPMark Verified | Agents and tools | 9 of 9 | Nothing to fix | ||
| McStasBench | Math and science | 10 of 10 | Nothing to fix | ||
| MeasEval | Math and science | 8 of 8 | Nothing to fix | ||
| MeasureBench | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| MECCANO Action Recognition | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| MecEng | Coding | 8 of 8 | Nothing to fix | ||
| Med-EASi | Professional work | 7 of 7 | Nothing to fix | ||
| MedCalc-Bench (original) | Professional work | 9 of 9 | Nothing to fix | ||
| MedCalc-Bench Verified | Professional work | 8 of 8 | Nothing to fix | ||
| MedCode | Professional work | 11 of 11 | Nothing to fix | ||
| MedDialog | Professional work | 7 of 7 | Nothing to fix | ||
| MedGenVidQA 2026 Visual Answer Localization | Professional work | 8 of 8 | Nothing to fix | ||
| Media Bias Identification Benchmark | Safety and security | 8 of 8 | Nothing to fix | ||
| Medical Bench | Professional work | 9 of 9 | Nothing to fix | ||
| MedICaT | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Medicinal Chemistry (Anthropic) | Math and science | 5 of 5 | Nothing to fix | ||
| MedPhysBench | Professional work | 10 of 10 | Nothing to fix | ||
| MedProcNER | Professional work | 7 of 7 | Nothing to fix | ||
| MedQA | Professional work | 9 of 9 | Nothing to fix | ||
| MedQA Taiwan | Professional work | 3 of 3 | Nothing to fix | ||
| MedQA USMLE (5 options) | Professional work | 3 of 3 | Nothing to fix | ||
| MedQuAD | Professional work | 7 of 7 | Nothing to fix | ||
| MedReadMe | Professional work | 7 of 7 | Nothing to fix | ||
| MedSafe-Dx v0.3 | Professional work | 10 of 10 | Nothing to fix | ||
| MedScribe | Professional work | 10 of 10 | Nothing to fix | ||
| MedSecId | Professional work | 8 of 8 | Nothing to fix | ||
| MedVidQA 2023 Medical Instructional Question Generation | Professional work | 7 of 7 | Nothing to fix | ||
| MedXpertQA | Professional work | 10 of 10 | Nothing to fix | ||
| MedXpertQA MM Extended | Professional work | 11 of 11 | Nothing to fix | ||
| MedXpertQA-MM | Professional work | 10 of 10 | Nothing to fix | ||
| Meeseeks Utility Rate | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MELD | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| MemGUI-Bench | Agents and tools | 10 of 10 | Nothing to fix | ||
| Memory Discipline | Knowledge and language | 9 of 9 | Nothing to fix | ||
| MemoryBench (Meta) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| MemoryTrap Bench: PHP arrays, Go maps | Coding | 9 of 9 | Nothing to fix | ||
| MemTrapBench | Knowledge and language | 7 of 7 | Nothing to fix | ||
| MeQSum | Professional work | 8 of 8 | Nothing to fix | ||
| MER-UniBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MerchantBench | Professional work | 9 of 9 | Nothing to fix | ||
| Meta action-belief coherence probe | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta agentic cyber misuse refusal | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta agentic evaluations grounded in user traffic | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta Agents Research Environments (ARE) / Gaia2 | Agents and tools | 10 of 10 | Nothing to fix | ||
| Meta BioDesign Proxy | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta BioProtocol Proxy | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta catastrophic cyber misuse refusal | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta chemical-agent refusal evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta Code-Fetch Under User Requests | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta curated CTF set | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta cyber false-refusal benchmark | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta Cyber Misuse Chat | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta cyber misuse request set | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta holistic adversarial audit with Petri Bloom | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta instructed selective performance evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta internal agent hallucination evaluation | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Meta internal sycophancy evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta prompted evaluation awareness | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta targeted agentic evaluations | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta training-stage provenance probe | Safety and security | 4 of 4 | Nothing to fix | ||
| Meta Underspecified Intent evaluation | Knowledge and language | 5 of 5 | Nothing to fix | ||
| Meta-Album | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Meta-Dataset | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| META-GUI | Agents and tools | 7 of 7 | Nothing to fix | ||
| Meta-World | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| MetagenomicsBench | Safety and security | 11 of 11 | Nothing to fix | ||
| METR Capability Index | Coding | 10 of 10 | Nothing to fix | ||
| METR Time Horizon 1.1 | Coding | 8 of 8 | Nothing to fix | ||
| METR Time Horizon 2 | Coding | 10 of 10 | Nothing to fix | ||
| MetroLLM-Bench | Agents and tools | 9 of 9 | Nothing to fix | ||
| MeViS | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| MEWC | Professional work | 4 of 4 | Nothing to fix | ||
| MEWC v2 | Professional work | 4 of 4 | Nothing to fix | ||
| MGSM | Knowledge and language | 10 of 10 | Nothing to fix | ||
| MGSM-Rev2 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MIA-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Microsoft internal multimodal RAI benchmark | Safety and security | 3 of 3 | Nothing to fix | ||
| Microsoft RAI benchmark: Grounding | Knowledge and language | 3 of 3 | Nothing to fix | ||
| Microsoft RAI benchmark: Harmful Content Continuation (DR-3) | Safety and security | 4 of 4 | Nothing to fix | ||
| Microsoft RAI benchmark: Harmful Content Summarization (DR-3) | Safety and security | 4 of 4 | Nothing to fix | ||
| Microsoft RAI benchmark: Jailbreak (DR-1) | Safety and security | 4 of 4 | Nothing to fix | ||
| Microsoft RAI benchmark: Third Party Harm (DR-1) | Safety and security | 4 of 4 | Nothing to fix | ||
| Microsoft RAI benchmark: Ungroundedness | Knowledge and language | 3 of 3 | Nothing to fix | ||
| MicroVQA | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Miðeind Icelandic ARC-Challenge | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Miðeind Icelandic Case Declension | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Miðeind Icelandic Case Government | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Miðeind Icelandic Generative Quality | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Miðeind Icelandic Grammar Correction | Knowledge and language | 11 of 11 | Nothing to fix | ||
| Miðeind Icelandic Grammatical Error Detection | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Miðeind Icelandic Idioms | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Miðeind Icelandic Popular Culture | Knowledge and language | 11 of 11 | Nothing to fix | ||
| Miðeind Icelandic Terminology | Knowledge and language | 11 of 11 | Nothing to fix | ||
| MIEB | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MIEB Lite | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MILU | Knowledge and language | 10 of 10 | Nothing to fix | ||
| MIMIC-CXR (classification) | Professional work | 7 of 7 | Nothing to fix | ||
| MIMIC-CXR report generation | Professional work | 7 of 7 | Nothing to fix | ||
| Mind2Web | Agents and tools | 4 of 4 | Nothing to fix | ||
| Mind2Web-SC | Agents and tools | 8 of 8 | Nothing to fix | ||
| MindCube | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| MInDS-14 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| MineBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Minecraft Builder Action Prediction | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| MINERVA | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| miniF2F | Math and science | 5 of 5 | Nothing to fix | ||
| Minimal-LinuxBench (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| MiniMax TTS Multilingual Test Set | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| MIRA Bench | Professional work | 5 of 5 | Nothing to fix | ||
| MIREX 2026 Music Evaluation via CMI-RewardBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MIREX 2026 Music Performance Difficulty Prediction | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| MirrorDojo | Safety and security | 4 of 4 | Nothing to fix | ||
| MirrorDyn | Safety and security | 4 of 4 | Nothing to fix | ||
| Misaligned Behavior in Realistic Work Environments | Safety and security | 5 of 5 | Nothing to fix | ||
| Misbehavior (Thinking Machines) | Safety and security | 4 of 4 | Nothing to fix | ||
| Misc CTF challenges (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| MISHAP-Bench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| MissionReport | Professional work | 11 of 11 | Nothing to fix | ||
| Mistral 175 unsafe prompts | Safety and security | 3 of 3 | Nothing to fix | ||
| Mistral internal function calling benchmark | Agents and tools | 3 of 3 | Nothing to fix | ||
| MIT-Adobe FiveK | Agents and tools | 8 of 8 | Nothing to fix | ||
| Mitosis Bench | Professional work | 11 of 11 | Nothing to fix | ||
| MIVAS Bench | Agents and tools | 10 of 10 | Nothing to fix | ||
| MKA | Professional work | 5 of 5 | Nothing to fix | ||
| MKQA | Knowledge and language | 7 of 7 | Nothing to fix | ||
| ML-Dev-Bench | Coding | 7 of 7 | Nothing to fix | ||
| MLCommons AILuminate | Safety and security | 9 of 9 | Nothing to fix | ||
| MLE-bench | Coding | 9 of 9 | Nothing to fix | ||
| MLS-Bench | Coding | 11 of 11 | Nothing to fix | ||
| MLVU | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| MM-FinEval | Professional work | 8 of 8 | Nothing to fix | ||
| MMAR | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| mMARCO | Knowledge and language | 7 of 7 | Nothing to fix | ||
| MMArt-Bench | Agents and tools | 8 of 8 | Nothing to fix | ||
| MMAU | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MMAU-Pro | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MMBench-CN v1.1 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MMDM | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MMDocIR | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MMEB | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| MMEB-V2 | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| MMEnterprise-Bench | Professional work | 10 of 10 | Nothing to fix | ||
| MMGenre | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MMIE | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| MMLongBench | Knowledge and language | 5 of 5 | Nothing to fix | ||
| MMLongBench-Doc | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| MMLU Anatomy | Professional work | 3 of 3 | Nothing to fix | ||
| MMLU calibration (expected calibration error) | Knowledge and language | 3 of 3 | Nothing to fix | ||
| MMLU Clinical Knowledge | Professional work | 3 of 3 | Nothing to fix | ||
| MMLU College Biology | Knowledge and language | 3 of 3 | Nothing to fix | ||
| MMLU Medical Genetics | Professional work | 3 of 3 | Nothing to fix | ||
| MMLU Professional Medicine | Professional work | 3 of 3 | Nothing to fix | ||
| MMLU-Pro | Knowledge and language | 9 of 9 | Nothing to fix | ||
| MMLU-ProX | Knowledge and language | 9 of 9 | Nothing to fix | ||
| MMLU-Redux | Knowledge and language | 5 of 5 | Nothing to fix | ||
| MMMB | Knowledge and language | 5 of 5 | Nothing to fix | ||
| MMMLU | Knowledge and language | 10 of 10 | Nothing to fix | ||
| MMMU | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MMMU-Pro | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| MMMU-Pro Extended | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| MMOU | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MMSearch-Plus | Agents and tools | 10 of 10 | Nothing to fix | ||
| MMSI-Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MMSI-Video-Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MMStar | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MMSU | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| MMVU | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Mobile-Eval-E | Agents and tools | 8 of 8 | Nothing to fix | ||
| Mobile-Eval-v2 | Agents and tools | 7 of 7 | Nothing to fix | ||
| MobilePA-Bench | Agents and tools | 10 of 10 | Nothing to fix | ||
| MobileWorld | Agents and tools | 10 of 10 | Nothing to fix | ||
| Model Spec Conversation Replay | Safety and security | 4 of 4 | Nothing to fix | ||
| Model Spec Targeted Cases | Safety and security | 4 of 4 | Nothing to fix | ||
| ModelBench: Reliability Under Constraints | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Mol-Instructions | Safety and security | 5 of 5 | Nothing to fix | ||
| Mol-Instructions Chemical Entity Recognition | Safety and security | 8 of 8 | Nothing to fix | ||
| Mol-Instructions Chemical-disease Interaction Extraction | Safety and security | 8 of 8 | Nothing to fix | ||
| Mol-Instructions Chemical-protein Interaction Extraction | Safety and security | 8 of 8 | Nothing to fix | ||
| Mol-Instructions Description-guided Molecule Design | Safety and security | 8 of 8 | Nothing to fix | ||
| Mol-Instructions Forward Reaction Prediction | Safety and security | 8 of 8 | Nothing to fix | ||
| Mol-Instructions Multi-choice | Safety and security | 8 of 8 | Nothing to fix | ||
| Mol-Instructions Reagent Prediction | Safety and security | 8 of 8 | Nothing to fix | ||
| Mol-Instructions Retrosynthesis | Safety and security | 8 of 8 | Nothing to fix | ||
| Mol-Instructions True or False | Safety and security | 8 of 8 | Nothing to fix | ||
| MolecularIQ | Safety and security | 10 of 10 | Nothing to fix | ||
| MOLTRON LLM Showdown v1.0 | Agents and tools | 9 of 9 | Nothing to fix | ||
| MOLTRON LLM Showdown v2.0 | Agents and tools | 9 of 9 | Nothing to fix | ||
| MOLTRON LLM Showdown v3.0 | Agents and tools | 10 of 10 | Nothing to fix | ||
| Moment finding (Google DeepMind) | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Moneda AdviceBench | Professional work | 7 of 7 | Nothing to fix | ||
| Moneda CiteBench | Professional work | 7 of 7 | Nothing to fix | ||
| Moneda FilingsBench | Professional work | 7 of 7 | Nothing to fix | ||
| Moneda GroundBench | Professional work | 7 of 7 | Nothing to fix | ||
| Moneda PortfolioBench | Professional work | 7 of 7 | Nothing to fix | ||
| Money talks | Safety and security | 3 of 3 | Nothing to fix | ||
| Monitorability Sandbagging | Safety and security | 5 of 5 | Nothing to fix | ||
| MonitorBench | Safety and security | 8 of 8 | Nothing to fix | ||
| Monorepo-Bench | Coding | 6 of 6 | Nothing to fix | ||
| Monthly-SWEBench | Coding | 10 of 10 | Nothing to fix | ||
| Morphology-to-molecule matching (Anthropic) | Math and science | 5 of 5 | Nothing to fix | ||
| MortgageTax | Professional work | 10 of 10 | Nothing to fix | ||
| MOSE | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MOSEv2 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MotionBench | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| MP-DocVQA | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MQuAKE | Knowledge and language | 7 of 7 | Nothing to fix | ||
| MRCR v2 | Knowledge and language | 5 of 5 | Nothing to fix | ||
| mRobust04 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| MS MARCO Question Answering | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MS-CXR-T | Professional work | 7 of 7 | Nothing to fix | ||
| MS² | Professional work | 7 of 7 | Nothing to fix | ||
| MSD-I Multimodal Genre Classification | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MSEB | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| MSQA | Knowledge and language | 5 of 5 | Nothing to fix | ||
| MT-AIME (Cohere internal) | Knowledge and language | 3 of 3 | Nothing to fix | ||
| MTEB | Other | 10 of 10 | Nothing to fix | ||
| MTEB Arena | Other | 8 of 8 | Nothing to fix | ||
| MTHv2 | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| MTOB | Knowledge and language | 4 of 4 | Nothing to fix | ||
| MTOP | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MTSamples clinical entity recognition | Professional work | 7 of 7 | Nothing to fix | ||
| MTVQA-Test | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MU-Bench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| MuChoMusic | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MuirBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Multi Agent Coordination Bench | Agents and tools | 9 of 9 | Nothing to fix | ||
| Multi-Agent Arena Codenames | Agents and tools | 10 of 10 | Nothing to fix | ||
| Multi-Agent Arena Diplomacy | Agents and tools | 11 of 11 | Nothing to fix | ||
| Multi-Agent Arena Risk | Agents and tools | 11 of 11 | Nothing to fix | ||
| Multi-Agent Arena Social Poker | Agents and tools | 10 of 10 | Nothing to fix | ||
| Multi-agent search (Anthropic internal benchmark) | Agents and tools | 4 of 4 | Nothing to fix | ||
| Multi-dialect SQL evaluation (Cohere) | Coding | 4 of 4 | Nothing to fix | ||
| Multi-factual (Anthropic) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Multi-hop search factuality | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Multi-IF | Knowledge and language | 5 of 5 | Nothing to fix | ||
| Multi-LexSum | Professional work | 8 of 8 | Nothing to fix | ||
| Multi-Scenario Anomaly Detection | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| MultiBLiMP | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MultiChallenge Audio (OpenAI) | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| MultiChallenge Audio (Xiaomi) | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| MultiClinSum | Professional work | 8 of 8 | Nothing to fix | ||
| MultiEURLEX | Professional work | 9 of 9 | Nothing to fix | ||
| Multilingual Entity Linking of Occupations | Professional work | 8 of 8 | Nothing to fix | ||
| Multilingual MMLU (Meta) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Multilingual Understanding | Knowledge and language | 3 of 3 | Nothing to fix | ||
| MultiLoKo | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Multimodal policy red-teaming (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| Multimodal RewardBench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Multimodal Troubleshooting Virology | Safety and security | 7 of 7 | Nothing to fix | ||
| MultiNERD | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MultiNRC | Knowledge and language | 10 of 10 | Nothing to fix | ||
| MultiPL-E | Coding | 8 of 8 | Nothing to fix | ||
| MultiRC | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MultiRef-Compass | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Multiturn Jailbreak Evaluations | Safety and security | 5 of 5 | Nothing to fix | ||
| MuMiN | Knowledge and language | 7 of 7 | Nothing to fix | ||
| MUSDB-ALT | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MUSEBENCH | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MUSIC Instrument Separation | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| MUSIC-21 Instrument Separation | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| MusicEval | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| MuSR | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MusTBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MuTual | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MVBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| MWAHAHA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| MyPCBench | Agents and tools | 10 of 10 | Nothing to fix | ||
| MysteryMechanism | Math and science | 11 of 11 | Nothing to fix | ||
| Mythos | Knowledge and language | 7 of 7 | Nothing to fix | ||
| NAFD E1 Financial Reporting | Professional work | 9 of 9 | Nothing to fix | ||
| NaijaRC | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Nakdimon Modern Hebrew Test Set | Knowledge and language | 8 of 8 | Nothing to fix | ||
| NanoGPT Speedrun | Coding | 5 of 5 | Nothing to fix | ||
| NarrativeBench | Knowledge and language | 9 of 9 | Nothing to fix | ||
| NASA-QA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Native Code Exploitation (Meta) | Safety and security | 4 of 4 | Nothing to fix | ||
| NativeLongBench | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Natural Adversarial Objects | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| NATURAL PLAN | Agents and tools | 8 of 8 | Nothing to fix | ||
| Natural World Tasks (NeWT 2021) | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Natural2Code | Coding | 5 of 5 | Nothing to fix | ||
| NaturalProofs | Math and science | 8 of 8 | Nothing to fix | ||
| NautData test set | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| NAVI | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Navital Gym | Agents and tools | 9 of 9 | Nothing to fix | ||
| Needle In A Haystack | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Nejumi LLM Leaderboard | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Nejumi LLM Leaderboard 3 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Neptune | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Network CTF challenges (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| Neutrality evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| NevIR | Knowledge and language | 8 of 8 | Nothing to fix | ||
| NewsGuard false audio claims audit | Knowledge and language | 7 of 7 | Nothing to fix | ||
| NewsQA | Knowledge and language | 7 of 7 | Nothing to fix | ||
| NExT-QA | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Next.js Agent Evals | Coding | 10 of 10 | Nothing to fix | ||
| NIFTY Market Movement | Professional work | 8 of 8 | Nothing to fix | ||
| Nigerian Pidgin ASR | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| NL2Repo-Bench | Coding | 10 of 10 | Nothing to fix | ||
| NLEBench | Knowledge and language | 8 of 8 | Nothing to fix | ||
| NLPCC 2024 Chinese Essay Discourse Logic Evaluation and Integration | Knowledge and language | 8 of 8 | Nothing to fix | ||
| NLVR2 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| No-CoT math time horizon | Safety and security | 4 of 4 | Nothing to fix | ||
| NoAdivines: Safe Order-Taking in Informal Spanish | Knowledge and language | 9 of 9 | Nothing to fix | ||
| nocaps | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| nocot-bench (NCRI) | Safety and security | 11 of 11 | Nothing to fix | ||
| NoLiMa | Knowledge and language | 8 of 8 | Nothing to fix | ||
| NollySenti | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Non-native Children's Automatic Speech Assessment Challenge | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| NorQuAD | Knowledge and language | 8 of 8 | Nothing to fix | ||
| NorSID | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Norwegian Courts Bitext Mining | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Norwegian Parliamentary Speech Corpus | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| NOVA-63 | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Novel compiler task (Anthropic) | Coding | 5 of 5 | Nothing to fix | ||
| Novel logical reasoning problems expert evaluation (Z.ai) | Math and science | 3 of 3 | Nothing to fix | ||
| NoveltyBench | Knowledge and language | 9 of 9 | Nothing to fix | ||
| NTCIR-18 HIDDEN-RAD | Professional work | 7 of 7 | Nothing to fix | ||
| NTCIR-18 MedNLP-CHAT | Professional work | 8 of 8 | Nothing to fix | ||
| Nufo | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Numerical results extraction from randomized controlled trials | Professional work | 10 of 10 | Nothing to fix | ||
| NusaWrites | Knowledge and language | 8 of 8 | Nothing to fix | ||
| NusaX | Knowledge and language | 8 of 8 | Nothing to fix | ||
| nuScenes Detection | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| NWPU-RESISC45 | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| o11y-bench | Agents and tools | 10 of 10 | Nothing to fix | ||
| OAI-repo Sabotage v2 | Safety and security | 4 of 4 | Nothing to fix | ||
| OASIS (agent safety) | Safety and security | 8 of 8 | Nothing to fix | ||
| Object Detection Leaderboard | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Objectron | Vision, audio and robotics | 3 of 3 | Nothing to fix | ||
| OCBench | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Occluded Video Instance Segmentation | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| OCNLI | Knowledge and language | 8 of 8 | Nothing to fix | ||
| OCR-Reasoning | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| OCRBench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| OCRBench v2 | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| ODinW-13 | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| OdysseysMega | Agents and tools | 9 of 9 | Nothing to fix | ||
| OfficePrism | Professional work | 10 of 10 | Nothing to fix | ||
| OfficeQA | Professional work | 6 of 6 | Nothing to fix | ||
| OfficeQA Pro | Professional work | 11 of 11 | Nothing to fix | ||
| OJBench | Coding | 11 of 11 | Nothing to fix | ||
| Okapi ARC | Math and science | 7 of 7 | Nothing to fix | ||
| Okapi Bench: API Freshness | Coding | 11 of 11 | Nothing to fix | ||
| Okapi HellaSwag | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Olam Labs Multi-Agent Arena | Agents and tools | 10 of 10 | Nothing to fix | ||
| olmOCR-Bench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| OmicsBench v1 | Safety and security | 10 of 10 | Nothing to fix | ||
| Omni3D | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| OmniACT | Agents and tools | 8 of 8 | Nothing to fix | ||
| OmniCap-IF | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| OmniDocBench v1.5 | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| OmniDocBench v1.6 | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| OmniGAIA | Agents and tools | 11 of 11 | Nothing to fix | ||
| OmniLabel | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Omnilingua-AudioMC | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Omnilingual-Gaia2 | Knowledge and language | 5 of 5 | Nothing to fix | ||
| OmniParseBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| OmniSpatial | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| OmniVBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| OmniVideoBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| OncologyBench | Safety and security | 11 of 11 | Nothing to fix | ||
| Online-Mind2Web | Agents and tools | 10 of 10 | Nothing to fix | ||
| OOD-Speech | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| OOLONG-Pairs | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Open Arabic LLM Leaderboard v1 | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Open Ended Questions (hallucination) | Knowledge and language | 3 of 3 | Nothing to fix | ||
| Open Ko-LLM Leaderboard | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Open LLM Leaderboard v1 | Other | 8 of 8 | Nothing to fix | ||
| Open Tabletop GM Narrative Quality Probe | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Open TTS Leaderboard 1.0.0 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Open TTS Leaderboard 1.1.0 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Open TTS Leaderboard 1.2.0 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Open TTS Leaderboard 1.3.0 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Open TTS Leaderboard 1.4.0 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Open TTS Leaderboard 1.5.0 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| OpenAI Epoch Capabilities Index (OECI) | Coding | 4 of 4 | Nothing to fix | ||
| OpenAI monitorability evals | Safety and security | 4 of 4 | Nothing to fix | ||
| OpenAI MRCR | Knowledge and language | 5 of 5 | Nothing to fix | ||
| OpenAI MRCR v2 | Knowledge and language | 9 of 9 | Nothing to fix | ||
| OpenAI PRs | Coding | 5 of 5 | Nothing to fix | ||
| OpenAI Research Engineer Interview: Coding | Coding | 4 of 4 | Nothing to fix | ||
| OpenAI Research Engineer Interview: Multiple Choice | Coding | 5 of 5 | Nothing to fix | ||
| OpenAI-Proof Q&A | Coding | 5 of 5 | Nothing to fix | ||
| OpenAudioBench | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| OpenBookQA | Knowledge and language | 9 of 9 | Nothing to fix | ||
| OpenBounds V2 Pilot | Coding | 9 of 9 | Nothing to fix | ||
| OpenCanopy | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| OpenFrontiers-Bench | Math and science | 10 of 10 | Nothing to fix | ||
| OpenGameEval | Coding | 9 of 9 | Nothing to fix | ||
| OpenGameEval Debug Evals | Coding | 10 of 10 | Nothing to fix | ||
| OpenGameEval Expanded Eval Set | Coding | 10 of 10 | Nothing to fix | ||
| Opening Lines Arena | Knowledge and language | 10 of 10 | Nothing to fix | ||
| OpenLegalRetrieval-DE | Professional work | 9 of 9 | Nothing to fix | ||
| OpenLex3D | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| OpenRSI Index ACE Playbook Inspection and Repair | Coding | 8 of 8 | Nothing to fix | ||
| OpenRSI Index Isaac Lab PegInsert Reward Search | Coding | 8 of 8 | Nothing to fix | ||
| OpenRSI Index Kev Decision Architecture | Coding | 8 of 8 | Nothing to fix | ||
| OpenRSI Index Learnability-Aware Long/Short CoT Adaptation | Coding | 9 of 9 | Nothing to fix | ||
| OpenRSI Index Marin Scaling Ladder | Coding | 8 of 8 | Nothing to fix | ||
| OpenRSI Index MInference 32-Head Sparse Prefill Kernel | Coding | 8 of 8 | Nothing to fix | ||
| OpenRSI Index Molmo2 Video-Pointing Inference Strategy | Coding | 8 of 8 | Nothing to fix | ||
| OpenRSI Index MolmoWeb Interaction Context Allocation | Coding | 8 of 8 | Nothing to fix | ||
| OpenRSI Index ReasonIR Difficulty Curriculum | Coding | 8 of 8 | Nothing to fix | ||
| OpenRSI Index SGLang Omni TTS Runtime | Coding | 8 of 8 | Nothing to fix | ||
| OpenStudio-MCP LLM Test Benchmark v1.2.1 | Agents and tools | 8 of 8 | Nothing to fix | ||
| OpenSUN3D Open-Vocabulary 3D Object Instance Search | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Operator prompt injection scenarios | Safety and security | 3 of 3 | Nothing to fix | ||
| Opportunistic sandbox escape (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| OptimBench Dispatch | Professional work | 9 of 9 | Nothing to fix | ||
| Opus Magnum Bench | Agents and tools | 9 of 9 | Nothing to fix | ||
| OPUS-100 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| OR-Bench | Safety and security | 9 of 9 | Nothing to fix | ||
| ORAV Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Organic Chemistry (Anthropic) | Math and science | 5 of 5 | Nothing to fix | ||
| Organic Chemistry V2 (Anthropic) | Math and science | 5 of 5 | Nothing to fix | ||
| OSI-Bench | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| OSWorld | Agents and tools | 10 of 10 | Nothing to fix | ||
| OSWorld 2.0 | Agents and tools | 11 of 11 | Nothing to fix | ||
| OSWorld 2.0 (Aug 2026) | Agents and tools | 10 of 10 | Nothing to fix | ||
| OSWorld 2.1 | Agents and tools | 10 of 10 | Nothing to fix | ||
| OSWorld 2.1 (offline subset) | Agents and tools | 9 of 9 | Nothing to fix | ||
| OSWorld-G | Agents and tools | 10 of 10 | Nothing to fix | ||
| OSWorld-Verified | Agents and tools | 10 of 10 | Nothing to fix | ||
| OTBench | Coding | 8 of 8 | Nothing to fix | ||
| OutboundEval | Knowledge and language | 8 of 8 | Nothing to fix | ||
| OVBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| OVEN | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Over-eager behavior on GUI computer use tasks (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| Overconfidence: verifying command-line syntax (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| Overstepping boundaries on ambiguous and high-risk tasks | Safety and security | 5 of 5 | Nothing to fix | ||
| OVO-Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| OWE-Bench | Coding | 11 of 11 | Nothing to fix | ||
| Òwe-Voice | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| P3D-Bench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Package registry publishing evaluation (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| PACT | Agents and tools | 9 of 9 | Nothing to fix | ||
| PAI-Bench | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| PaintBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| PaintSkills | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| PAIRS (Parallel Images) | Safety and security | 8 of 8 | Nothing to fix | ||
| Pairwise Bioweaponization Campaign | Safety and security | 4 of 4 | Nothing to fix | ||
| Pakistan Local Context LLM Suite | Knowledge and language | 9 of 9 | Nothing to fix | ||
| PalmX 2025 General Culture | Knowledge and language | 8 of 8 | Nothing to fix | ||
| PalmX 2025 Islamic Culture | Knowledge and language | 8 of 8 | Nothing to fix | ||
| PAM-Audio | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Panda-CVL | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Pantograph frontier-LLM robot benchmark (Pandroid) | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| PaperBench | Coding | 9 of 9 | Nothing to fix | ||
| PaperBenchX | Math and science | 10 of 10 | Nothing to fix | ||
| Paragraph Instruction Following | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Parallel Search Capability Leaderboard | Other | 10 of 10 | Nothing to fix | ||
| ParallelKernelBench | Coding | 10 of 10 | Nothing to fix | ||
| Parametric CAD Bench | Professional work | 10 of 10 | Nothing to fix | ||
| Paraphrase Stability Benchmark | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Parashift Document AI Benchmark | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| ParaShoot | Knowledge and language | 7 of 7 | Nothing to fix | ||
| PARCEL | Professional work | 8 of 8 | Nothing to fix | ||
| ParseBench | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| ParseBench-2 | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| ParsiNLU | Knowledge and language | 8 of 8 | Nothing to fix | ||
| ParsiNLU Mizan Literary Translation | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Participation in junk science (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| PartiPrompts | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Passkey Retrieval | Knowledge and language | 3 of 3 | Nothing to fix | ||
| Passport Control | Knowledge and language | 10 of 10 | Nothing to fix | ||
| PatchCamelyon | Professional work | 10 of 10 | Nothing to fix | ||
| PatchEval | Safety and security | 10 of 10 | Nothing to fix | ||
| PatchEval-Verified | Safety and security | 10 of 10 | Nothing to fix | ||
| PatientAgentBench | Professional work | 9 of 9 | Nothing to fix | ||
| PAWS-X | Knowledge and language | 7 of 7 | Nothing to fix | ||
| PCBBench (pcbbench.org) | Professional work | 10 of 10 | Nothing to fix | ||
| pCLUE | Knowledge and language | 8 of 8 | Nothing to fix | ||
| PCT (Polarized Contrastive Pairs) | Safety and security | 10 of 10 | Nothing to fix | ||
| PDAGENT-BENCH | Professional work | 10 of 10 | Nothing to fix | ||
| PE Civil Bench | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Pedagogy Benchmark | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Penn epilepsy clinic notes | Professional work | 7 of 7 | Nothing to fix | ||
| People’s Daily and Children’s Fairy Tale | Knowledge and language | 7 of 7 | Nothing to fix | ||
| PerceptionBench (Moonshot) | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| PerceptionComp | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| Perceptual Attacks on Embodied AI (Robocurve) | Safety and security | 10 of 10 | Nothing to fix | ||
| PerCoR | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Periapical Lesions Bench | Professional work | 11 of 11 | Nothing to fix | ||
| Persian IFEval | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Personal Superintelligence Index (Meta) | Other | 4 of 4 | Nothing to fix | ||
| Personal-device assistant prompt injection eval | Safety and security | 4 of 4 | Nothing to fix | ||
| PersonQA | Knowledge and language | 4 of 4 | Nothing to fix | ||
| PerspectiveGap | Knowledge and language | 10 of 10 | Nothing to fix | ||
| PestCLEF 2026 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Petri Bloom | Safety and security | 4 of 4 | Nothing to fix | ||
| Phage-plasmid Co-evolution | Safety and security | 6 of 6 | Nothing to fix | ||
| Phare | Knowledge and language | 10 of 10 | Nothing to fix | ||
| PHEE | Professional work | 7 of 7 | Nothing to fix | ||
| PHINC | Knowledge and language | 7 of 7 | Nothing to fix | ||
| PhoATIS | Knowledge and language | 7 of 7 | Nothing to fix | ||
| PhoMT | Knowledge and language | 8 of 8 | Nothing to fix | ||
| PhoneShell iOS Agent Benchmark | Agents and tools | 9 of 9 | Nothing to fix | ||
| PHYBench | Math and science | 10 of 10 | Nothing to fix | ||
| Phylogenetics (Anthropic) | Math and science | 4 of 4 | Nothing to fix | ||
| PhysClosureBench | Math and science | 9 of 9 | Nothing to fix | ||
| Physical-Engineering's Last Exam | Professional work | 10 of 10 | Nothing to fix | ||
| PhysicianBench | Professional work | 10 of 10 | Nothing to fix | ||
| PHYSICS | Math and science | 9 of 9 | Nothing to fix | ||
| Physics-IQ | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Physics-IQ Verified | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| PHYSICS-Verified | Math and science | 9 of 9 | Nothing to fix | ||
| PhysicsFinals | Math and science | 4 of 4 | Nothing to fix | ||
| PhyX | Math and science | 10 of 10 | Nothing to fix | ||
| PianoBench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| PICABench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| PiiClustering | Professional work | 8 of 8 | Nothing to fix | ||
| Pile-test | Other | 3 of 3 | Nothing to fix | ||
| PinchBench 2.0 | Agents and tools | 9 of 9 | Nothing to fix | ||
| Ping or Peek | Safety and security | 11 of 11 | Nothing to fix | ||
| Pipecat Speech-to-Text Benchmark | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| PIQA | Knowledge and language | 9 of 9 | Nothing to fix | ||
| PiRhDy Accompaniment Suggestion | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| PitVQA | Professional work | 7 of 7 | Nothing to fix | ||
| Pixel2Perspective | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| PixMo-Count | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Places Audio Captions (Hindi) | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Plain English Summarization of Contracts | Professional work | 8 of 8 | Nothing to fix | ||
| Plain Language Adaptation of Biomedical Abstracts | Professional work | 8 of 8 | Nothing to fix | ||
| PLawBench | Professional work | 10 of 10 | Nothing to fix | ||
| PlayWorld | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| PLCWorld | Professional work | 10 of 10 | Nothing to fix | ||
| PlotPoints Round 2: Multi-Turn Arena | Knowledge and language | 8 of 8 | Nothing to fix | ||
| PlotPoints Round 3: Multi-Turn Arena | Knowledge and language | 9 of 9 | Nothing to fix | ||
| PlotPoints Round 4: Willingness and Judgment | Safety and security | 10 of 10 | Nothing to fix | ||
| PlugInject | Safety and security | 3 of 3 | Nothing to fix | ||
| PMC-VQA | Professional work | 10 of 10 | Nothing to fix | ||
| PNCExtract | Math and science | 8 of 8 | Nothing to fix | ||
| PoemSum | Knowledge and language | 7 of 7 | Nothing to fix | ||
| PokéBench: Raw Vision + Buttons | Agents and tools | 10 of 10 | Nothing to fix | ||
| PokéBench: Tool Assisted Benchmark | Agents and tools | 10 of 10 | Nothing to fix | ||
| Pokémon VGC Bench (Ricotoma) | Agents and tools | 10 of 10 | Nothing to fix | ||
| PolEval 2025 Gender-inclusive Proofreading | Knowledge and language | 7 of 7 | Nothing to fix | ||
| PolEval 2025 Gender-sensitive Translation | Knowledge and language | 7 of 7 | Nothing to fix | ||
| PolicyBench | Professional work | 10 of 10 | Nothing to fix | ||
| Polish Speech Emotion Recognition Challenge | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Political Persuasion Parallel Generation | Safety and security | 4 of 4 | Nothing to fix | ||
| Political Science LLM Benchmark | Other | 11 of 11 | Nothing to fix | ||
| Poly-Guard Bench | Safety and security | 9 of 9 | Nothing to fix | ||
| PolyMath | Math and science | 10 of 10 | Nothing to fix | ||
| POPE | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Post-OCR Text Correction in Sanskrit | Knowledge and language | 8 of 8 | Nothing to fix | ||
| PosterBench | Professional work | 8 of 8 | Nothing to fix | ||
| PostTrainBench | Coding | 10 of 10 | Nothing to fix | ||
| PostTrainBench 1.1 | Coding | 10 of 10 | Nothing to fix | ||
| PostTrainBench 1.2 | Coding | 11 of 11 | Nothing to fix | ||
| PowerBench | Agents and tools | 10 of 10 | Nothing to fix | ||
| PPR10K | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| PPTGen-Bench | Professional work | 9 of 9 | Nothing to fix | ||
| PRBench | Professional work | 11 of 11 | Nothing to fix | ||
| Pre-Flight | Professional work | 11 of 11 | Nothing to fix | ||
| Precedent vs. Spec | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Preference on ChatGPT and API prompts | Knowledge and language | 4 of 4 | Nothing to fix | ||
| PReMind-Bench | Safety and security | 9 of 9 | Nothing to fix | ||
| PresentBench | Professional work | 10 of 10 | Nothing to fix | ||
| Pressure-Tested Alignment | Safety and security | 9 of 9 | Nothing to fix | ||
| PrimeVul | Safety and security | 8 of 8 | Nothing to fix | ||
| prinzbench | Professional work | 9 of 9 | Nothing to fix | ||
| PRISM-Bench (audio and video generation) | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| PrivacyLens | Safety and security | 9 of 9 | Nothing to fix | ||
| Privy Synthetic PII Protocol Trace Dataset | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Proactive refusals of high-risk tasks | Safety and security | 3 of 3 | Nothing to fix | ||
| Proactive Searches | Safety and security | 4 of 4 | Nothing to fix | ||
| ProactiveVideoQA | Vision, audio and robotics | 3 of 3 | Nothing to fix | ||
| ProbSum 2023 | Professional work | 8 of 8 | Nothing to fix | ||
| ProcBench | Math and science | 8 of 8 | Nothing to fix | ||
| ProcessBench | Math and science | 8 of 8 | Nothing to fix | ||
| ProdBench (Tencent) | Agents and tools | 5 of 5 | Nothing to fix | ||
| Production Deception - Adversarial | Safety and security | 3 of 3 | Nothing to fix | ||
| Production Jailbreaks | Safety and security | 4 of 4 | Nothing to fix | ||
| ProfBench | Professional work | 10 of 10 | Nothing to fix | ||
| ProgramDistill | Coding | 10 of 10 | Nothing to fix | ||
| Progress classification (Google DeepMind) | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| Project Euler (MathArena) | Math and science | 7 of 7 | Nothing to fix | ||
| Prompt injection attacks in search and function calling | Safety and security | 4 of 4 | Nothing to fix | ||
| Prompt Injection in Benchmark Environments | Safety and security | 4 of 4 | Nothing to fix | ||
| Prompt Injection in Model-Generated Environments | Safety and security | 4 of 4 | Nothing to fix | ||
| Prompt injection live bug bounty (Anthropic with Gray Swan) | Safety and security | 4 of 4 | Nothing to fix | ||
| Prompt injection robustness | Safety and security | 5 of 5 | Nothing to fix | ||
| Prompt injection robustness against adaptive attackers, browser use (Anthropic) | Safety and security | 11 of 11 | Nothing to fix | ||
| Prompt injection robustness against adaptive attackers, coding (Anthropic) | Safety and security | 6 of 6 | Nothing to fix | ||
| Prompt injection robustness against adaptive attackers, computer use (Anthropic) | Safety and security | 6 of 6 | Nothing to fix | ||
| Prompt injection, general tool use | Safety and security | 4 of 4 | Nothing to fix | ||
| Prompt injection, turn injections in tool use | Safety and security | 4 of 4 | Nothing to fix | ||
| Prompt to Product | Coding | 10 of 10 | Nothing to fix | ||
| Prompt-to-Bench HO-16 | Professional work | 9 of 9 | Nothing to fix | ||
| Prompt-to-Bench-16 | Professional work | 9 of 9 | Nothing to fix | ||
| ProofBench | Math and science | 10 of 10 | Nothing to fix | ||
| ProofBench v1.1 | Math and science | 10 of 10 | Nothing to fix | ||
| Prophet Arena | Math and science | 10 of 10 | Nothing to fix | ||
| Protein Design (Anthropic) | Math and science | 5 of 5 | Nothing to fix | ||
| ProteinGym | Math and science | 9 of 9 | Nothing to fix | ||
| ProtocolQA Open-Ended | Safety and security | 6 of 6 | Nothing to fix | ||
| Protocols (Anthropic life sciences) | Safety and security | 5 of 5 | Nothing to fix | ||
| PST50 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| PsyCoMark: Conspiracy Detection | Knowledge and language | 8 of 8 | Nothing to fix | ||
| PsyCoMark: Conspiracy Marker Extraction | Knowledge and language | 8 of 8 | Nothing to fix | ||
| PUBHEALTH | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Public Benefits Bench | Professional work | 11 of 11 | Nothing to fix | ||
| PubMed causal language use | Professional work | 7 of 7 | Nothing to fix | ||
| PUCRS inpatient fall detection | Professional work | 7 of 7 | Nothing to fix | ||
| PulseBench-Tab | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| PUMaVOS | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| PushWorld | Agents and tools | 7 of 7 | Nothing to fix | ||
| Putnam-like | Math and science | 8 of 8 | Nothing to fix | ||
| PutnamBench | Math and science | 10 of 10 | Nothing to fix | ||
| PVD Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Pwn CTF challenges (Anthropic) | Safety and security | 5 of 5 | Nothing to fix | ||
| PWNBench v0.1 | Safety and security | 10 of 10 | Nothing to fix | ||
| PWS-Decisions | Agents and tools | 8 of 8 | Nothing to fix | ||
| PWS-Finance | Professional work | 8 of 8 | Nothing to fix | ||
| PWS-Health | Professional work | 8 of 8 | Nothing to fix | ||
| PWS-Legal | Professional work | 8 of 8 | Nothing to fix | ||
| pyATS-Bench | Coding | 10 of 10 | Nothing to fix | ||
| Q-Bench Image-Pair Perception | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Q-Bench Single-Image Perception | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Q2D-Chunk | Knowledge and language | 8 of 8 | Nothing to fix | ||
| QAG-360K | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| QAMPARI | Knowledge and language | 8 of 8 | Nothing to fix | ||
| QASC | Math and science | 7 of 7 | Nothing to fix | ||
| QBQTC | Knowledge and language | 7 of 7 | Nothing to fix | ||
| QEVD-FIT-COACH | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| QIAS 2025 Islamic Inheritance Reasoning | Professional work | 8 of 8 | Nothing to fix | ||
| QIAS 2025 Islamic Knowledge Assessment | Knowledge and language | 8 of 8 | Nothing to fix | ||
| QIMMA | Knowledge and language | 10 of 10 | Nothing to fix | ||
| QMSum | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Qodo AI Code Review Benchmark | Coding | 8 of 8 | Nothing to fix | ||
| Quadruped reinforcement learning task (Anthropic) | Coding | 5 of 5 | Nothing to fix | ||
| QuALITY | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Quant-Arena | Professional work | 10 of 10 | Nothing to fix | ||
| QuantBench | Math and science | 4 of 4 | Nothing to fix | ||
| Quartets Audio-to-Score Transcription | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| QuartetsMIREX 2026 Audio-to-Score Transcription | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Questflow Finance Benchmark | Professional work | 9 of 9 | Nothing to fix | ||
| Qur’an QA 2023 Passage Retrieval | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Qur’an QA 2023 Reading Comprehension | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Qwen Chinese human evaluation | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Qwen-Audio-Agent Smart Cockpit Benchmark | Agents and tools | 8 of 8 | Nothing to fix | ||
| QwenClawBench | Agents and tools | 11 of 11 | Nothing to fix | ||
| QwenQoderBench | Coding | 11 of 11 | Nothing to fix | ||
| QwenReactBench | Coding | 10 of 10 | Nothing to fix | ||
| QwenSVGBench | Coding | 10 of 10 | Nothing to fix | ||
| QwenSWEBench | Coding | 10 of 10 | Nothing to fix | ||
| QwenWebBench | Coding | 3 of 3 | Nothing to fix | ||
| RACE | Knowledge and language | 9 of 9 | Nothing to fix | ||
| RACE-H | Knowledge and language | 4 of 4 | Nothing to fix | ||
| RACE-M | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Radiological and Nuclear Expert Knowledge | Safety and security | 4 of 4 | Nothing to fix | ||
| Radiology's Last Exam (RadLE) | Professional work | 9 of 9 | Nothing to fix | ||
| RadLE 2.0 | Professional work | 9 of 9 | Nothing to fix | ||
| RadNLP 2024 Lung Cancer Staging | Professional work | 8 of 8 | Nothing to fix | ||
| RadNLP 2024 Radiology Report Segmentation | Professional work | 7 of 7 | Nothing to fix | ||
| RadSum23 | Professional work | 7 of 7 | Nothing to fix | ||
| RAGBench | Knowledge and language | 7 of 7 | Nothing to fix | ||
| RAGTruth | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Ramen Bench | Coding | 10 of 10 | Nothing to fix | ||
| Ramp Accounting Bench | Professional work | 11 of 11 | Nothing to fix | ||
| Ramp SWE-Bench v1 | Coding | 9 of 9 | Nothing to fix | ||
| Ramp SWE-Bench v2 | Coding | 11 of 11 | Nothing to fix | ||
| Random Complex Tasks (LongCat) | Agents and tools | 4 of 4 | Nothing to fix | ||
| RandomBench | Math and science | 10 of 10 | Nothing to fix | ||
| Rapidata Multi-Step Image Editing (Identity Drift) | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Ravnursson Faroese Speech and Transcripts | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Razavi-Bench | Professional work | 10 of 10 | Nothing to fix | ||
| RBench | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| Re:Baltica Latvian and Russian chatbot sources | Knowledge and language | 8 of 8 | Nothing to fix | ||
| React Native Evals | Coding | 10 of 10 | Nothing to fix | ||
| ReactBench | Coding | 10 of 10 | Nothing to fix | ||
| ReactionBench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Read My Aura | Safety and security | 9 of 9 | Nothing to fix | ||
| Readability (Perplexity) | Knowledge and language | 3 of 3 | Nothing to fix | ||
| Real World VoiceEQ | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Real-SWE | Coding | 10 of 10 | Nothing to fix | ||
| Real-SWE 1.1 | Coding | 11 of 11 | Nothing to fix | ||
| Real-time Audio-Visual Interaction Evaluation | Vision, audio and robotics | 3 of 3 | Nothing to fix | ||
| Real-World Audio Recording | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| Real-World Finance (Anthropic) | Professional work | 5 of 5 | Nothing to fix | ||
| Real-World Finance v2 (Anthropic) | Professional work | 4 of 4 | Nothing to fix | ||
| Real-world vulnerability discovery | Safety and security | 4 of 4 | Nothing to fix | ||
| Real2Sim | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Real5-OmniDocBench | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| RealCommBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| RealCompanion | Knowledge and language | 9 of 9 | Nothing to fix | ||
| realitybench | Coding | 9 of 9 | Nothing to fix | ||
| Realtime-Bench | Agents and tools | 11 of 11 | Nothing to fix | ||
| RealWorkDecision | Professional work | 11 of 11 | Nothing to fix | ||
| RealWorldQA | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Reasoning Dial | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Reasoning Gym | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Reasoning-Bias-Reporting-Quality | Safety and security | 10 of 10 | Nothing to fix | ||
| ReasonSeg | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ReasonVOS | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Recipe1M image and recipe retrieval | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Recognition of model-written user turns (Anthropic) | Other | 3 of 3 | Nothing to fix | ||
| RECON | Agents and tools | 11 of 11 | Nothing to fix | ||
| ReCoRD | Knowledge and language | 8 of 8 | Nothing to fix | ||
| RecreationBench | Agents and tools | 10 of 10 | Nothing to fix | ||
| Red Team Defense Simulator | Safety and security | 7 of 7 | Nothing to fix | ||
| Red Teaming Network pairwise safety comparison | Safety and security | 4 of 4 | Nothing to fix | ||
| Red-Pen Bench | Coding | 8 of 8 | Nothing to fix | ||
| RedCode | Safety and security | 4 of 4 | Nothing to fix | ||
| Reddit-Impacts | Professional work | 7 of 7 | Nothing to fix | ||
| RedditImpacts 2.0 | Professional work | 8 of 8 | Nothing to fix | ||
| RedlineBench | Professional work | 9 of 9 | Nothing to fix | ||
| RedstoneBench | Agents and tools | 10 of 10 | Nothing to fix | ||
| Redtape | Professional work | 8 of 8 | Nothing to fix | ||
| Ref-DAVIS17 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Ref-L4 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Ref-ZOM | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| RefAdv-S | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| RefAV | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| RefCOCO / RefCOCO+ / RefCOCOg | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| RefSpatial-Bench | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| RefSpatial-Expand-Bench | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| RefSpatial-Expand-Bench Location | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| RefSpatial-Expand-Bench Placement | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Refusal to assist with AI safety R&D (Anthropic) | Safety and security | 5 of 5 | Nothing to fix | ||
| RegexEval | Coding | 7 of 7 | Nothing to fix | ||
| RELATE | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| RelyToolBench | Agents and tools | 10 of 10 | Nothing to fix | ||
| Remote Labor Index | Professional work | 12 of 12 | Nothing to fix | ||
| ReMuQ | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| RepCount-A | Vision, audio and robotics | 3 of 3 | Nothing to fix | ||
| RepoBench | Coding | 4 of 4 | Nothing to fix | ||
| RepoQA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Research Plan Generation | Coding | 7 of 7 | Nothing to fix | ||
| ResearchClawBench | Math and science | 9 of 9 | Nothing to fix | ||
| ResearchRubrics | Agents and tools | 10 of 10 | Nothing to fix | ||
| Resolve | Coding | 7 of 7 | Nothing to fix | ||
| Respecting Auto-Review | Safety and security | 6 of 6 | Nothing to fix | ||
| Respecting Warnings | Safety and security | 6 of 6 | Nothing to fix | ||
| RESPIN-S1.0 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Retrobench v0.4 | Coding | 10 of 10 | Nothing to fix | ||
| Rev CTF challenges (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| RevBench (Polyvor Labs) | Safety and security | 9 of 9 | Nothing to fix | ||
| Revisited Oxford and Paris | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Reward-hack-prone coding tasks (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| RewardBench | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Rewind-attacker cyber adversarial robustness (Anthropic) | Safety and security | 6 of 6 | Nothing to fix | ||
| RH SWE-bench v1 | Coding | 9 of 9 | Nothing to fix | ||
| RICH | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Riemann-bench | Math and science | 12 of 12 | Nothing to fix | ||
| Right Answer, Wrong Reason: ML Judgment Calls | Coding | 9 of 9 | Nothing to fix | ||
| RISE | Agents and tools | 4 of 4 | Nothing to fix | ||
| RISEBench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| RiskRubric | Safety and security | 10 of 10 | Nothing to fix | ||
| RivaBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| RL rollout audit for beyond-episode drives | Safety and security | 4 of 4 | Nothing to fix | ||
| RLBench | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| RMS-AQA | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| RMU:ECHR | Professional work | 7 of 7 | Nothing to fix | ||
| RNG-Bench | Agents and tools | 10 of 10 | Nothing to fix | ||
| RoadmapBench | Coding | 10 of 10 | Nothing to fix | ||
| Robo-bench | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| RoboCasa | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| RoboCasa GR1 Tabletop Tasks | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| RoboChallenge | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| RoboColosseum | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Robocurve simple tasks (LLM control of YAM arms) | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| RoboDojo | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| RoboDojo-RC Tier 1 | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| RoboLab | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| Robots.txt Benchmark | Coding | 9 of 9 | Nothing to fix | ||
| RoboTwin 1.0 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| RobotWorld | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| RoboWorld | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Robust IF Bench | Knowledge and language | 5 of 5 | Nothing to fix | ||
| Robust Multi-view Depth | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| RoleBench | Knowledge and language | 4 of 4 | Nothing to fix | ||
| RoleMRC | Knowledge and language | 5 of 5 | Nothing to fix | ||
| RollerCoasterTycoonBench | Agents and tools | 8 of 8 | Nothing to fix | ||
| ROOMELSA | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Rotom Champ Bench | Agents and tools | 10 of 10 | Nothing to fix | ||
| Rotom Gen9 Bench | Agents and tools | 9 of 9 | Nothing to fix | ||
| RoughCut-Bench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| RP2K | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| RRG24 | Professional work | 8 of 8 | Nothing to fix | ||
| RRN | Knowledge and language | 7 of 7 | Nothing to fix | ||
| RSI Arena | Coding | 9 of 9 | Nothing to fix | ||
| RSI Bench | Coding | 10 of 10 | Nothing to fix | ||
| RSI Control Arena | Safety and security | 9 of 9 | Nothing to fix | ||
| RSI-Exam | Coding | 11 of 11 | Nothing to fix | ||
| RSI-Inference Bench | Coding | 10 of 10 | Nothing to fix | ||
| RSICD | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| RSIEval | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| RSLiveBench-UAV | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| RSNA Knee AUC10 | Professional work | 10 of 10 | Nothing to fix | ||
| RSNA Knee LLM Label Extraction | Professional work | 10 of 10 | Nothing to fix | ||
| rTaste | Coding | 10 of 10 | Nothing to fix | ||
| RTE | Knowledge and language | 8 of 8 | Nothing to fix | ||
| RTLLM | Professional work | 7 of 7 | Nothing to fix | ||
| RTLScout Bench (suite-v2) | Professional work | 9 of 9 | Nothing to fix | ||
| Rule2DRC | Professional work | 10 of 10 | Nothing to fix | ||
| RULER | Knowledge and language | 10 of 10 | Nothing to fix | ||
| RULERv2 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| RuLES | Knowledge and language | 9 of 9 | Nothing to fix | ||
| RuMedBench | Professional work | 7 of 7 | Nothing to fix | ||
| RuneBench | Agents and tools | 10 of 10 | Nothing to fix | ||
| RVG-Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| RWE-bench | Professional work | 11 of 11 | Nothing to fix | ||
| RWSearch | Agents and tools | 11 of 11 | Nothing to fix | ||
| S-Eval | Safety and security | 8 of 8 | Nothing to fix | ||
| SA-37 interactive segmentation | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| SaaS-Bench | Agents and tools | 10 of 10 | Nothing to fix | ||
| SAFE-Teen | Safety and security | 10 of 10 | Nothing to fix | ||
| SAFE-Teen Health advice | Safety and security | 10 of 10 | Nothing to fix | ||
| SAFE-Teen Relationship boundaries | Safety and security | 10 of 10 | Nothing to fix | ||
| SAFE-Teen Urgent safety | Safety and security | 10 of 10 | Nothing to fix | ||
| Safety (MiniMax in-house) | Safety and security | 3 of 3 | Nothing to fix | ||
| Safety-Scene Vision Benchmark | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| SafetyBench | Safety and security | 3 of 3 | Nothing to fix | ||
| SAGE (Vals) | Other | 11 of 11 | Nothing to fix | ||
| SAGE v1.1 (Vals) | Other | 11 of 11 | Nothing to fix | ||
| Sakana kana-letter reading-order set | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SALAD-Bench | Safety and security | 8 of 8 | Nothing to fix | ||
| SAM Audio-Bench | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Samrómur Children | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Samskritam Bench v0 | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Samskritam Bench v1 | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Sandbox Bench | Safety and security | 7 of 7 | Nothing to fix | ||
| Sandbox escape evaluation (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| SandhiKosh | Knowledge and language | 7 of 7 | Nothing to fix | ||
| SATBench | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Sayfullina Soft Skill Span Extraction | Professional work | 7 of 7 | Nothing to fix | ||
| SB10k | Knowledge and language | 7 of 7 | Nothing to fix | ||
| ScaLA | Knowledge and language | 10 of 10 | Nothing to fix | ||
| ScanNet-1500 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ScanNet++ v2 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ScarfBench | Coding | 8 of 8 | Nothing to fix | ||
| scBench (original study) | Safety and security | 9 of 9 | Nothing to fix | ||
| scBench-Long | Safety and security | 11 of 11 | Nothing to fix | ||
| scBench-Long (original study) | Safety and security | 9 of 9 | Nothing to fix | ||
| SceneFun3D | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| ScholarCatalyst | Math and science | 10 of 10 | Nothing to fix | ||
| SCI-AGENT-EXAM | Math and science | 9 of 9 | Nothing to fix | ||
| SciCap | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| SciCode | Coding | 10 of 10 | Nothing to fix | ||
| SciDocs | Math and science | 8 of 8 | Nothing to fix | ||
| Science's Last Exam | Math and science | 9 of 9 | Nothing to fix | ||
| ScienceQA | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ScienceVision | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| ScienceWorld | Math and science | 8 of 8 | Nothing to fix | ||
| Scientists' First Exam (SFE) | Vision, audio and robotics | 6 of 6 | Nothing to fix | ||
| SciTaste-Bench | Math and science | 9 of 9 | Nothing to fix | ||
| SciUniverse | Math and science | 10 of 10 | Nothing to fix | ||
| SciVid | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ScreenSpot | Agents and tools | 8 of 8 | Nothing to fix | ||
| ScreenSpot-Pro | Agents and tools | 10 of 10 | Nothing to fix | ||
| ScreenSpot-v2 | Agents and tools | 9 of 9 | Nothing to fix | ||
| SEA-HELM | Knowledge and language | 10 of 10 | Nothing to fix | ||
| SEA-HELM Audio | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Search triggering | Agents and tools | 4 of 4 | Nothing to fix | ||
| SEC-bench Pro | Safety and security | 10 of 10 | Nothing to fix | ||
| SecCodeBench | Safety and security | 10 of 10 | Nothing to fix | ||
| SecIT Bench | Safety and security | 11 of 11 | Nothing to fix | ||
| SecQA | Safety and security | 8 of 8 | Nothing to fix | ||
| SecRespond | Safety and security | 10 of 10 | Nothing to fix | ||
| Secure and defensive coding (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| Seed-TTS-eval | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| Seed1.5-VL internal benchmark (ByteDance Seed in-house) | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| SeedClawBench | Agents and tools | 11 of 11 | Nothing to fix | ||
| SeedKernelBench | Coding | 4 of 4 | Nothing to fix | ||
| Segment Anything Video | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Segment Anything with Concepts | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Select machine learning engineering tasks from METR | Coding | 3 of 3 | Nothing to fix | ||
| Self-preference bias (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| Semantic DLP: Catch the Data Leak | Knowledge and language | 7 of 7 | Nothing to fix | ||
| SemaPLC Function Track | Professional work | 10 of 10 | Nothing to fix | ||
| SemaPLC Project-Context Track | Professional work | 9 of 9 | Nothing to fix | ||
| SemEval 2023 Task 3: Genre, Framing and Persuasion Techniques in Online News | Knowledge and language | 7 of 7 | Nothing to fix | ||
| SemEval 2024 Task 1 semantic textual relatedness | Knowledge and language | 9 of 9 | Nothing to fix | ||
| SemEval 2026 Task 4: Narrative Story Similarity and Narrative Representation Learning | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SemEval-2015 Task 1 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| SEND Pedagogy Benchmark | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Senior Engineer Benchmark | Coding | 9 of 9 | Nothing to fix | ||
| Senior Engineer Benchmark (Every Checks) | Coding | 10 of 10 | Nothing to fix | ||
| Senior SWE-Bench | Coding | 11 of 11 | Nothing to fix | ||
| Sequitur Research Lab Automation | Safety and security | 4 of 4 | Nothing to fix | ||
| SettlerBench Season One | Agents and tools | 10 of 10 | Nothing to fix | ||
| SEvenLLM | Safety and security | 7 of 7 | Nothing to fix | ||
| Sexually Explicit Content Safety | Safety and security | 4 of 4 | Nothing to fix | ||
| SF20K Competition 2025 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SG-LegalClarify | Professional work | 10 of 10 | Nothing to fix | ||
| SGI-Bench | Math and science | 10 of 10 | Nothing to fix | ||
| SHADE-Arena | Safety and security | 6 of 6 | Nothing to fix | ||
| ShadowBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Shop Assistant Honesty Bench (UA/RU) | Agents and tools | 9 of 9 | Nothing to fix | ||
| Short Form Bench | Professional work | 10 of 10 | Nothing to fix | ||
| SHP2 Protein Function Prediction | Safety and security | 6 of 6 | Nothing to fix | ||
| SHROOM-Visions 2026 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SICK-NL | Knowledge and language | 7 of 7 | Nothing to fix | ||
| SIFo | Knowledge and language | 4 of 4 | Nothing to fix | ||
| SIGHUM Sanskrit Word Segmentation | Knowledge and language | 7 of 7 | Nothing to fix | ||
| SIGMORPHON 2023 Shared Task on Interlinear Glossing | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Silent copying of answers (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| SilverquiLLM-bench hob-medium | Coding | 10 of 10 | Nothing to fix | ||
| SIM-VAIL | Safety and security | 10 of 10 | Nothing to fix | ||
| SimpleBench | Math and science | 10 of 10 | Nothing to fix | ||
| SimpleQA | Knowledge and language | 9 of 9 | Nothing to fix | ||
| SimplerEnv | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| SimpleSafetyTests | Safety and security | 8 of 8 | Nothing to fix | ||
| SimpleVQA | Knowledge and language | 10 of 10 | Nothing to fix | ||
| SimReady Text-to-CAD | Professional work | 8 of 8 | Nothing to fix | ||
| SimReadyBench NV100 | Professional work | 10 of 10 | Nothing to fix | ||
| SimuVerity | Professional work | 9 of 9 | Nothing to fix | ||
| Single-turn benign request evaluations (Anthropic) | Safety and security | 6 of 6 | Nothing to fix | ||
| Single-turn harmful request evaluations (Anthropic) | Safety and security | 6 of 6 | Nothing to fix | ||
| SingMOS | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Sinhala Language Proficiency & Linguistic Mastery | Knowledge and language | 10 of 10 | Nothing to fix | ||
| SIQA | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Situational Awareness Dataset (SAD) | Safety and security | 10 of 10 | Nothing to fix | ||
| SkillInject | Safety and security | 4 of 4 | Nothing to fix | ||
| SkillsBench | Agents and tools | 10 of 10 | Nothing to fix | ||
| SkillSpan | Professional work | 7 of 7 | Nothing to fix | ||
| SkillSpan-ESCO | Professional work | 8 of 8 | Nothing to fix | ||
| skLEP | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SLAKE | Professional work | 5 of 5 | Nothing to fix | ||
| Slay the Spire Bench (alavico) | Agents and tools | 9 of 9 | Nothing to fix | ||
| Slide generation human evaluation (Z.ai) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| SlideVQA | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SLIP zero-shot transfer suite | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| SLoMO Audio Description Competition 2026 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SlopCodeBench | Coding | 11 of 11 | Nothing to fix | ||
| SLURP | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Small Model Agent Bench | Agents and tools | 9 of 9 | Nothing to fix | ||
| SMB Bench | Professional work | 10 of 10 | Nothing to fix | ||
| ŚMIGIEL | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SMILE Video Laugh Reasoning | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| SMM4H-HeaRD 2025 Detection of Insomnia in Clinical Notes | Professional work | 8 of 8 | Nothing to fix | ||
| SMM4H-HeaRD 2025 Multilingual Adverse Drug Event Detection | Professional work | 8 of 8 | Nothing to fix | ||
| SMM4H-HeaRD 2026 Detection of Insomnia in Clinical Notes | Professional work | 8 of 8 | Nothing to fix | ||
| SMM4H-HeaRD 2026 Flu Vaccine Effectiveness | Professional work | 8 of 8 | Nothing to fix | ||
| SMM4H-HeaRD 2026 MedSynth Dialogue-to-Note | Professional work | 8 of 8 | Nothing to fix | ||
| SMM4H-HeaRD 2026 Multilingual Adverse Drug Event Detection | Professional work | 8 of 8 | Nothing to fix | ||
| SMT 2025 | Math and science | 8 of 8 | Nothing to fix | ||
| SnakeBench | Agents and tools | 9 of 9 | Nothing to fix | ||
| Sneaky Sandbox Escape | Safety and security | 4 of 4 | Nothing to fix | ||
| Snorkel Agentic Coding | Coding | 9 of 9 | Nothing to fix | ||
| Snorkel Agentic Coding 2.0 | Coding | 11 of 11 | Nothing to fix | ||
| Snorkel Working Memory Test | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SnorkelFinance 2.0 | Professional work | 11 of 11 | Nothing to fix | ||
| SnorkelLegal | Professional work | 11 of 11 | Nothing to fix | ||
| SnorkelManufacturing | Professional work | 11 of 11 | Nothing to fix | ||
| SnorkelRevOps | Professional work | 11 of 11 | Nothing to fix | ||
| SnorkelSpatial | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SnorkelUnderwrite | Professional work | 8 of 8 | Nothing to fix | ||
| SnorkelUnderwrite 2.0 | Professional work | 11 of 11 | Nothing to fix | ||
| SnorkelWordle | Agents and tools | 8 of 8 | Nothing to fix | ||
| Snowflake AI_EXTRACT Document Extraction Evaluation | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| SO-Arena | Knowledge and language | 10 of 10 | Nothing to fix | ||
| SoccerNet-Caption | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Social Signal Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SOFC-Exp | Math and science | 8 of 8 | Nothing to fix | ||
| Solana Machine Trading | Professional work | 10 of 10 | Nothing to fix | ||
| SolidWorks Bench | Professional work | 10 of 10 | Nothing to fix | ||
| SongBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SongEval | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| SpaceChem Bench | Agents and tools | 10 of 10 | Nothing to fix | ||
| SPair-71k | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SPAR-Bench | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| SpartQA | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Spatial Agent Index v2 | Agents and tools | 10 of 10 | Nothing to fix | ||
| Spatial Intelligence Index v2 | Other | 10 of 10 | Nothing to fix | ||
| Spatial Pathfinding Reasoning Challenge | Agents and tools | 10 of 10 | Nothing to fix | ||
| Spatial Reasoning: Rotating ASCII Art | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| SpatialBench (LatchBio) | Math and science | 12 of 12 | Nothing to fix | ||
| SpatialBench-Long | Safety and security | 11 of 11 | Nothing to fix | ||
| SpatialBench-Long (original study) | Safety and security | 9 of 9 | Nothing to fix | ||
| SpatialSense VQA | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| SpatialTree | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| SPBench | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| Speech Agent Arena | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| Speech Robust Bench | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| SpeechEditBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SpeechIO ASR leaderboard | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SPEED-Bench | Other | 9 of 9 | Nothing to fix | ||
| SPGISpeech | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Spider | Coding | 9 of 9 | Nothing to fix | ||
| Spider 2.0 | Coding | 9 of 9 | Nothing to fix | ||
| Spiral-Bench v1.0 | Safety and security | 8 of 8 | Nothing to fix | ||
| SpireClimb StS1 A20 Heart | Agents and tools | 11 of 11 | Nothing to fix | ||
| SplatBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| SportsEval | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SpreadsheetBench | Professional work | 9 of 9 | Nothing to fix | ||
| SpreadsheetBench 2 | Professional work | 10 of 10 | Nothing to fix | ||
| SpreadsheetBench Verified | Professional work | 10 of 10 | Nothing to fix | ||
| SPY | Knowledge and language | 7 of 7 | Nothing to fix | ||
| SQL Reality Check | Coding | 8 of 8 | Nothing to fix | ||
| SQuAD | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SquadStack Conversational LLM Benchmark for Voice Agents | Agents and tools | 9 of 9 | Nothing to fix | ||
| SQuALITY | Knowledge and language | 4 of 4 | Nothing to fix | ||
| SREGym-Lite | Agents and tools | 9 of 9 | Nothing to fix | ||
| Sri Lankan local election chatbot audit | Knowledge and language | 7 of 7 | Nothing to fix | ||
| SSH stress-testing evaluation (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| Standard Refusal Evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| Stanford Online Products | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| STaRK | Knowledge and language | 8 of 8 | Nothing to fix | ||
| StarSkirmish Bench | Agents and tools | 11 of 11 | Nothing to fix | ||
| STATE-Bench Main 0.7.0 | Agents and tools | 8 of 8 | Nothing to fix | ||
| StationeryBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Stealth and Situational Awareness evaluations | Safety and security | 4 of 4 | Nothing to fix | ||
| STELLA-Bench | Safety and security | 9 of 9 | Nothing to fix | ||
| STEM Professor Bench | Math and science | 5 of 5 | Nothing to fix | ||
| StereoSet | Safety and security | 8 of 8 | Nothing to fix | ||
| STORAL | Knowledge and language | 7 of 7 | Nothing to fix | ||
| StoreBench | Professional work | 10 of 10 | Nothing to fix | ||
| StrataBench Land v1 | Professional work | 10 of 10 | Nothing to fix | ||
| StrataBench Land v1 Classify | Professional work | 9 of 9 | Nothing to fix | ||
| StrategyQA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| StreamingBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| StrongREJECT | Safety and security | 5 of 5 | Nothing to fix | ||
| StrongREJECT v2 (Meta) | Safety and security | 4 of 4 | Nothing to fix | ||
| StructEval | Coding | 10 of 10 | Nothing to fix | ||
| Structural Biology (Anthropic) | Math and science | 5 of 5 | Nothing to fix | ||
| Structured Translation (Cohere) | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Student-Work Grading | Professional work | 8 of 8 | Nothing to fix | ||
| StudentBench | Other | 9 of 9 | Nothing to fix | ||
| StudyArena Image | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Success detection (Google DeepMind) | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| sudo L7 | Coding | 12 of 12 | Nothing to fix | ||
| Sugar’s Kitchen AI Cooking Battle 17 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SugarCrepe | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Suicide & Self-Harm evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| Suicide Memory Bench | Safety and security | 10 of 10 | Nothing to fix | ||
| Sumi Japanese PII Corpus | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SummScreenFD | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SUMO Web Claims Summarization | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Supabase Evals | Agents and tools | 10 of 10 | Nothing to fix | ||
| SUPERB | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SUPERChem | Safety and security | 10 of 10 | Nothing to fix | ||
| SuperCLUE | Other | 9 of 9 | Nothing to fix | ||
| SuperCLUE Arena Image Editing | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SuperCLUE Arena Image to Video | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SuperCLUE Arena Reference to Video | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SuperCLUE Arena Text to Speech | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| SuperCLUE Arena Text to Video | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SuperCLUE Arena Web Coding | Coding | 9 of 9 | Nothing to fix | ||
| SuperGLUE | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SuperGPQA | Knowledge and language | 11 of 11 | Nothing to fix | ||
| SuperLim 2.0 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SuperSim | Knowledge and language | 7 of 7 | Nothing to fix | ||
| SURDS | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Surface Evolver Bench | Math and science | 9 of 9 | Nothing to fix | ||
| SVG-Bench (MiniMax) | Coding | 5 of 5 | Nothing to fix | ||
| SVG-Bench (StarVector) | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| SWAG | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Swarm Bench | Agents and tools | 5 of 5 | Nothing to fix | ||
| SWE Atlas | Coding | 10 of 10 | Nothing to fix | ||
| SWE Atlas Codebase QnA Extended | Coding | 11 of 11 | Nothing to fix | ||
| SWE Atlas Refactoring | Coding | 10 of 10 | Nothing to fix | ||
| SWE Atlas Test Writing | Coding | 10 of 10 | Nothing to fix | ||
| SWE Atlas v1.1 | Coding | 11 of 11 | Nothing to fix | ||
| SWE Refactor Bench | Coding | 10 of 10 | Nothing to fix | ||
| SWE-bench | Coding | 8 of 8 | Nothing to fix | ||
| SWE-bench CLI | Coding | 11 of 11 | Nothing to fix | ||
| SWE-bench Lite | Coding | 8 of 8 | Nothing to fix | ||
| SWE-bench Multimodal | Coding | 11 of 11 | Nothing to fix | ||
| SWE-bench Multimodal v2 | Coding | 11 of 11 | Nothing to fix | ||
| SWE-Bench Pro | Coding | 10 of 10 | Nothing to fix | ||
| SWE-Bench Pro Private | Coding | 10 of 10 | Nothing to fix | ||
| SWE-Bench Pro V2 | Coding | 10 of 10 | Nothing to fix | ||
| SWE-Bench Pro V2 Hard | Coding | 11 of 11 | Nothing to fix | ||
| SWE-bench Science | Coding | 10 of 10 | Nothing to fix | ||
| SWE-bench Verified (hard subset) | Coding | 5 of 5 | Nothing to fix | ||
| SWE-bench Verified Extended | Coding | 11 of 11 | Nothing to fix | ||
| SWE-bench-Live | Coding | 11 of 11 | Nothing to fix | ||
| SWE-Clar | Coding | 10 of 10 | Nothing to fix | ||
| SWE-CoWork | Coding | 10 of 10 | Nothing to fix | ||
| SWE-Dev | Coding | 4 of 4 | Nothing to fix | ||
| SWE-fficiency | Coding | 10 of 10 | Nothing to fix | ||
| SWE-Marathon | Coding | 11 of 11 | Nothing to fix | ||
| SWE-Marathon v1.1 | Coding | 11 of 11 | Nothing to fix | ||
| SWE-Pair | Coding | 8 of 8 | Nothing to fix | ||
| SWE-PI | Safety and security | 4 of 4 | Nothing to fix | ||
| SWE-Race | Coding | 9 of 9 | Nothing to fix | ||
| SWE-rebench | Coding | 10 of 10 | Nothing to fix | ||
| SWE-Serve | Coding | 10 of 10 | Nothing to fix | ||
| SWE-sweep | Coding | 11 of 11 | Nothing to fix | ||
| SweDiagnostics | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Swedish Analogy Test Set | Knowledge and language | 7 of 7 | Nothing to fix | ||
| SweFAQ 2.0 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SweNLI | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SweParaphrase 2.0 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SweSAT Synonyms 1.1 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| SweWiC | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SweWinogender 2.0 | Safety and security | 8 of 8 | Nothing to fix | ||
| SweWinograd 2.0 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Swiss-Judgment-Prediction | Professional work | 7 of 7 | Nothing to fix | ||
| SWT-bench | Coding | 8 of 8 | Nothing to fix | ||
| SycoBench-600 | Safety and security | 8 of 8 | Nothing to fix | ||
| Sycophancy evaluation (offline) | Safety and security | 3 of 3 | Nothing to fix | ||
| Sycophancy on user-provided prompts (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| SYNERGY systematic review screening | Professional work | 9 of 9 | Nothing to fix | ||
| SynLogic | Knowledge and language | 8 of 8 | Nothing to fix | ||
| SynthBench | Math and science | 9 of 9 | Nothing to fix | ||
| Synthetic Sheet Music Reasoning Benchmark | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| SysBench | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Szeged NER | Knowledge and language | 8 of 8 | Nothing to fix | ||
| T-Eval | Agents and tools | 8 of 8 | Nothing to fix | ||
| T2I-NSI | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| T2LSC-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| TabFact | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Table Judge Benchmark | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Tacit Knowledge and Troubleshooting | Safety and security | 5 of 5 | Nothing to fix | ||
| Tacit Knowledge Brainstorm (Open-Ended) | Safety and security | 4 of 4 | Nothing to fix | ||
| TacMind-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Taiwan Mandarin Speech with Video: Speech Enhancement | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Taiwan National Dental Technician Licensing Examination 2023 | Professional work | 7 of 7 | Nothing to fix | ||
| TakeoffBench-v1 | Professional work | 10 of 10 | Nothing to fix | ||
| TalentCLEF 2025 Job Title-Based Skill Prediction | Professional work | 8 of 8 | Nothing to fix | ||
| TalentCLEF 2025 Multilingual Job Title Matching | Professional work | 8 of 8 | Nothing to fix | ||
| TalentCLEF 2026 Contextualized Job-Person Matching | Professional work | 8 of 8 | Nothing to fix | ||
| TalentCLEF 2026 Job-Skill Matching with Skill Type Classification | Professional work | 8 of 8 | Nothing to fix | ||
| TallyQA | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| TAP-Vid | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Taste-Bench | Coding | 10 of 10 | Nothing to fix | ||
| TasteVal | Coding | 10 of 10 | Nothing to fix | ||
| TAT-QA | Professional work | 8 of 8 | Nothing to fix | ||
| Tatoeba sentence retrieval | Knowledge and language | 9 of 9 | Nothing to fix | ||
| tau-bench | Agents and tools | 9 of 9 | Nothing to fix | ||
| tau2-bench | Agents and tools | 10 of 10 | Nothing to fix | ||
| tau2-Noise (LongCat) | Agents and tools | 4 of 4 | Nothing to fix | ||
| tau3-bench | Agents and tools | 9 of 9 | Nothing to fix | ||
| TavernBench | Agents and tools | 9 of 9 | Nothing to fix | ||
| Tavily Company Search Benchmark | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Tax Agent Bench | Professional work | 12 of 12 | Nothing to fix | ||
| TaxaBench-8k | Safety and security | 7 of 7 | Nothing to fix | ||
| TaxCalcBench | Professional work | 9 of 9 | Nothing to fix | ||
| TaxCalcBench v2 | Professional work | 10 of 10 | Nothing to fix | ||
| TaxEval v2 | Professional work | 10 of 10 | Nothing to fix | ||
| TCSExam | Coding | 9 of 9 | Nothing to fix | ||
| TCSExam-Lite | Coding | 9 of 9 | Nothing to fix | ||
| TE-ca | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Teaching Monster Challenge | Professional work | 7 of 7 | Nothing to fix | ||
| TeleQnA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| TeluguMixBench | Knowledge and language | 8 of 8 | Nothing to fix | ||
| TempCompass | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Template Following generalization (Qwen) | Agents and tools | 3 of 3 | Nothing to fix | ||
| Temporal reasoning with search | Knowledge and language | 4 of 4 | Nothing to fix | ||
| Tencent WorkBuddy Bench | Coding | 10 of 10 | Nothing to fix | ||
| Terminal-Bench 2.0 | Agents and tools | 10 of 10 | Nothing to fix | ||
| Terminal-Bench 2.1 | Agents and tools | 10 of 10 | Nothing to fix | ||
| Terminal-Bench 2.1 Extended | Agents and tools | 11 of 11 | Nothing to fix | ||
| Terminal-Bench 3.0 | Agents and tools | 11 of 11 | Nothing to fix | ||
| Terminal-Bench 4.1 | Agents and tools | 12 of 12 | Nothing to fix | ||
| Terminal-Bench Hard | Agents and tools | 6 of 6 | Nothing to fix | ||
| Terminal-Bench Math Modeling | Coding | 9 of 9 | Nothing to fix | ||
| Terminal-Bench-Science | Math and science | 11 of 11 | Nothing to fix | ||
| Testing model's emotional inteligence | Knowledge and language | 9 of 9 | Nothing to fix | ||
| TestPrism | Coding | 10 of 10 | Nothing to fix | ||
| TetrisBench | Agents and tools | 9 of 9 | Nothing to fix | ||
| Text-based reinforcement learning task (Anthropic) | Coding | 4 of 4 | Nothing to fix | ||
| TextEdit | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| TextEdit Real | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| TextEdit Virtual | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| TextVQA | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| TGIF-QA | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| The Aggregate Chart Recall | Knowledge and language | 10 of 10 | Nothing to fix | ||
| The Aggregate Lichess Pairwise Strength | Agents and tools | 10 of 10 | Nothing to fix | ||
| The Aggregate Lichess Rating Prediction | Agents and tools | 10 of 10 | Nothing to fix | ||
| The Aggregate Personal Style Imitation | Knowledge and language | 11 of 11 | Nothing to fix | ||
| The Met Dataset | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| The Pushback Test | Professional work | 8 of 8 | Nothing to fix | ||
| TheGoldman Retrieval | Professional work | 7 of 7 | Nothing to fix | ||
| TheOne Routing Decisions | Knowledge and language | 7 of 7 | Nothing to fix | ||
| TheoPhysEval | Math and science | 9 of 9 | Nothing to fix | ||
| ThermoQA | Math and science | 8 of 8 | Nothing to fix | ||
| ThinkFollow (Qwen) | Knowledge and language | 3 of 3 | Nothing to fix | ||
| ThinkGeo | Agents and tools | 10 of 10 | Nothing to fix | ||
| Thrive Holdings helpdesk ticket resolution benchmark | Professional work | 11 of 11 | Nothing to fix | ||
| TIME (time series forecasting) | Math and science | 8 of 8 | Nothing to fix | ||
| Time series forecasting task (Anthropic) | Coding | 5 of 5 | Nothing to fix | ||
| TimeLens-Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| TimelineBench | Professional work | 11 of 11 | Nothing to fix | ||
| TimeScope | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| TimeSpeak | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| TireSidewall-Bench: Multimodal Zero-Shot Tire OCR | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| TIS-100 Bench | Coding | 10 of 10 | Nothing to fix | ||
| TMHINT-QI | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| TMHINT-QI version II | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| TMMLU+ | Knowledge and language | 8 of 8 | Nothing to fix | ||
| TMMLU+ 1.1 | Knowledge and language | 9 of 9 | Nothing to fix | ||
| ToLD-Br | Safety and security | 8 of 8 | Nothing to fix | ||
| Toloka Arena Enterprise Tool Use (November 2026) | Agents and tools | 11 of 11 | Nothing to fix | ||
| Toloka Visual Question Answering Benchmark | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| TOMATO | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Tombo Full Transcript | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| TOMG-Bench | Safety and security | 10 of 10 | Nothing to fix | ||
| Tool use prompt injection evaluation (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| Tool-calling prompt injections | Safety and security | 3 of 3 | Nothing to fix | ||
| Toolathlon | Agents and tools | 11 of 11 | Nothing to fix | ||
| Toolathlon-Verified | Agents and tools | 12 of 12 | Nothing to fix | ||
| ToolCall-Badcase | Agents and tools | 3 of 3 | Nothing to fix | ||
| ToolRet | Agents and tools | 8 of 8 | Nothing to fix | ||
| ToolSandbox | Agents and tools | 10 of 10 | Nothing to fix | ||
| ToolUse (Qwen) | Agents and tools | 3 of 3 | Nothing to fix | ||
| TopiOCQA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| ToxicChat | Safety and security | 7 of 7 | Nothing to fix | ||
| ToxiGen (discriminative) | Safety and security | 3 of 3 | Nothing to fix | ||
| tr_gender_bias | Knowledge and language | 8 of 8 | Nothing to fix | ||
| TraceSpatial-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| TradeRank Arena | Professional work | 10 of 10 | Nothing to fix | ||
| Training distribution hack rate (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| Translation Human Scoring (Z.ai) | Knowledge and language | 3 of 3 | Nothing to fix | ||
| Travel Agent Compassion | Safety and security | 10 of 10 | Nothing to fix | ||
| TravelPlanner | Agents and tools | 8 of 8 | Nothing to fix | ||
| TreasuryQA Bench | Professional work | 11 of 11 | Nothing to fix | ||
| TREC 2017 LiveQA Medical Task | Professional work | 8 of 8 | Nothing to fix | ||
| TREC Million LLMs 2025 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| TREC Product Search 2024 | Knowledge and language | 7 of 7 | Nothing to fix | ||
| TREC Product Search 2025 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| TREC Related Product Recommendation 2025 | Knowledge and language | 8 of 8 | Nothing to fix | ||
| TreeBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| TREK-150 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| TrialDesignBench | Professional work | 9 of 9 | Nothing to fix | ||
| trivial-prompt-bench | Other | 9 of 9 | Nothing to fix | ||
| TriviaQA | Knowledge and language | 5 of 5 | Nothing to fix | ||
| TriWorldBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| TroubleshootingBench | Safety and security | 6 of 6 | Nothing to fix | ||
| TrustLLM JailbreakTrigger | Safety and security | 8 of 8 | Nothing to fix | ||
| TruthfulQA | Knowledge and language | 9 of 9 | Nothing to fix | ||
| ts-bench v1 | Coding | 8 of 8 | Nothing to fix | ||
| Tuesday Work Index | Professional work | 10 of 10 | Nothing to fix | ||
| TurBLiMP | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Turkish Dental Specialty Examination 2020 and 2021 | Professional work | 7 of 7 | Nothing to fix | ||
| Turn based Benchmarks | Agents and tools | 9 of 9 | Nothing to fix | ||
| TV-AD | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| TVBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| TVQA | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| tweetSentBR | Knowledge and language | 8 of 8 | Nothing to fix | ||
| Twitter URL Paraphrase Corpus | Knowledge and language | 8 of 8 | Nothing to fix | ||
| TxBench-Antibody-Discovery | Safety and security | 11 of 11 | Nothing to fix | ||
| TxBench-Oligo-Discovery | Safety and security | 11 of 11 | Nothing to fix | ||
| TxBench-Preclinical-Pharmacology | Safety and security | 11 of 11 | Nothing to fix | ||
| Typed Decision Bench | Professional work | 8 of 8 | Nothing to fix | ||
| U-Artifacts | Coding | 4 of 4 | Nothing to fix | ||
| U-MATH | Math and science | 8 of 8 | Nothing to fix | ||
| UA-Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| UCF101 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| UD Ancient Greek Perseus | Knowledge and language | 8 of 8 | Nothing to fix | ||
| UD Ancient Greek PROIEL | Knowledge and language | 8 of 8 | Nothing to fix | ||
| UD Latin CIRCSE | Knowledge and language | 8 of 8 | Nothing to fix | ||
| UD Latin ITTB | Knowledge and language | 8 of 8 | Nothing to fix | ||
| UD Latin Perseus | Knowledge and language | 8 of 8 | Nothing to fix | ||
| UD Latin PROIEL | Knowledge and language | 8 of 8 | Nothing to fix | ||
| UFO-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| UI-TARS 15 Games Collection | Agents and tools | 3 of 3 | Nothing to fix | ||
| UI-Vision Action Prediction | Agents and tools | 8 of 8 | Nothing to fix | ||
| UIT-ViCTSD | Knowledge and language | 8 of 8 | Nothing to fix | ||
| UK AISI and CAISI simulated enterprise network range | Safety and security | 9 of 9 | Nothing to fix | ||
| UK AISI correlated monitor failure | Safety and security | 4 of 4 | Nothing to fix | ||
| UK AISI CoT Controllability | Safety and security | 4 of 4 | Nothing to fix | ||
| UK AISI cyber ranges | Safety and security | 5 of 5 | Nothing to fix | ||
| UK AISI narrow cyber tasks | Safety and security | 4 of 4 | Nothing to fix | ||
| UK AISI no-CoT maths competition problems | Safety and security | 5 of 5 | Nothing to fix | ||
| UK AISI Out of Scope Supply Chain Attack | Safety and security | 4 of 4 | Nothing to fix | ||
| UK AISI sabotage continuation evaluation | Safety and security | 4 of 4 | Nothing to fix | ||
| UK AISI simulated sandbox escape monitor evasion | Safety and security | 4 of 4 | Nothing to fix | ||
| UK AISI unprompted sabotage evaluation | Safety and security | 5 of 5 | Nothing to fix | ||
| UK AISI vulnerability discovery challenges | Safety and security | 4 of 4 | Nothing to fix | ||
| Unauthorized Code Execution Bench | Safety and security | 4 of 4 | Nothing to fix | ||
| Unauthorized voice generation | Safety and security | 3 of 3 | Nothing to fix | ||
| Uncritically reporting flawed results (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| Unexpected container or code states | Safety and security | 3 of 3 | Nothing to fix | ||
| Ungrounded inference and sensitive trait attribution | Safety and security | 4 of 4 | Nothing to fix | ||
| UniClawBench | Agents and tools | 10 of 10 | Nothing to fix | ||
| Unicorn (Vision Safety) | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Unintended Engagement with External Agent Messages | Safety and security | 6 of 6 | Nothing to fix | ||
| UniREditBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| UnityBench-Mini3D | Coding | 10 of 10 | Nothing to fix | ||
| Urdu Language Speech Emotional Corpus | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| UrduSER | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| USAMO 2025 | Math and science | 9 of 9 | Nothing to fix | ||
| USAMO 2026 | Math and science | 9 of 9 | Nothing to fix | ||
| User Confirmations During Computer Use | Safety and security | 4 of 4 | Nothing to fix | ||
| UserBench (developer simulation) | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Uzbek Open ASR Leaderboard | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Uzbek Speech Corpus | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| V* | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Valerius Flaccus Intertextuality Benchmark | Knowledge and language | 8 of 8 | Nothing to fix | ||
| VALOR-32K | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Vals Finance Agent | Professional work | 11 of 11 | Nothing to fix | ||
| Vals Finance Agent v1 | Professional work | 10 of 10 | Nothing to fix | ||
| Vals Finance Agent v2 | Professional work | 12 of 12 | Nothing to fix | ||
| Vals Finance Agent v2.1 | Professional work | 12 of 12 | Nothing to fix | ||
| Vals Finance Index | Professional work | 12 of 12 | Nothing to fix | ||
| Vals Index v1 | Professional work | 10 of 10 | Nothing to fix | ||
| Vals Index v2 | Professional work | 11 of 11 | Nothing to fix | ||
| Vals Index v2.1 | Professional work | 12 of 12 | Nothing to fix | ||
| Vals Index v2.2 | Professional work | 12 of 12 | Nothing to fix | ||
| Vals Inference Provider Index | Professional work | 9 of 9 | Nothing to fix | ||
| Vals Legal Index | Professional work | 12 of 12 | Nothing to fix | ||
| Vals Multimodal Index | Other | 10 of 10 | Nothing to fix | ||
| Vals Multimodal Index v1 | Professional work | 9 of 9 | Nothing to fix | ||
| Vals Multimodal Index v1.1 | Professional work | 9 of 9 | Nothing to fix | ||
| Vals Router Index | Other | 8 of 8 | Nothing to fix | ||
| Vals RSI Index | Coding | 10 of 10 | Nothing to fix | ||
| Vals Science Index | Other | 12 of 12 | Nothing to fix | ||
| Vals Web Search Index | Agents and tools | 10 of 10 | Nothing to fix | ||
| VAmoS Bench | Agents and tools | 9 of 9 | Nothing to fix | ||
| VANTAGE-BENCH | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| VAREX-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| VariantBench | Safety and security | 11 of 11 | Nothing to fix | ||
| VBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| VBench image-to-video | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| VCBench | Professional work | 10 of 10 | Nothing to fix | ||
| VCIS | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| VCSUM | Knowledge and language | 3 of 3 | Nothing to fix | ||
| VCT-v2 | Safety and security | 12 of 12 | Nothing to fix | ||
| Vectara Hallucination Leaderboard (2025 dataset) | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Vending-Bench | Professional work | 9 of 9 | Nothing to fix | ||
| Vending-Bench 2 | Professional work | 10 of 10 | Nothing to fix | ||
| Vending-Bench Arena | Professional work | 10 of 10 | Nothing to fix | ||
| VERA-MH Suicide Risk | Safety and security | 9 of 9 | Nothing to fix | ||
| Verify Gym | Math and science | 9 of 9 | Nothing to fix | ||
| Verifying the Verifiers (Project APE CRED benchmark) | Math and science | 9 of 9 | Nothing to fix | ||
| VeriGap | Coding | 10 of 10 | Nothing to fix | ||
| VeriGen design problems | Professional work | 10 of 10 | Nothing to fix | ||
| VerilogEval v2 | Professional work | 9 of 9 | Nothing to fix | ||
| Veris FinAgent Arena v1.1 | Professional work | 10 of 10 | Nothing to fix | ||
| VerticalRTK | Agents and tools | 8 of 8 | Nothing to fix | ||
| Verus-Bench | Coding | 7 of 7 | Nothing to fix | ||
| VeruSAGE-Bench | Coding | 8 of 8 | Nothing to fix | ||
| VES Label Check | Math and science | 9 of 9 | Nothing to fix | ||
| VetEval | Professional work | 10 of 10 | Nothing to fix | ||
| VGI-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| VIALS | Safety and security | 10 of 10 | Nothing to fix | ||
| VIALS 2 | Safety and security | 11 of 11 | Nothing to fix | ||
| VIBE (MiniMax) | Coding | 9 of 9 | Nothing to fix | ||
| Vibe Code Bench 1-100 | Coding | 11 of 11 | Nothing to fix | ||
| Vibe Code Bench 1-100 v1.1 | Coding | 11 of 11 | Nothing to fix | ||
| Vibe Code Bench v1 | Coding | 9 of 9 | Nothing to fix | ||
| VIBE-Pro (MiniMax) | Coding | 4 of 4 | Nothing to fix | ||
| VIBE-V2 (MiniMax) | Coding | 5 of 5 | Nothing to fix | ||
| ViBench 2.0 | Coding | 11 of 11 | Nothing to fix | ||
| VibePrevisBench | Professional work | 10 of 10 | Nothing to fix | ||
| VibeVideoBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Video Object Segmentation under Transformations | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Video-ChatGPT video conversation benchmark | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Video-Holmes | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Video-Index | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| Video-MME | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Video-MMMU | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| VideoDR (with search) | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| VideoGAIA | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| VideoGen-RewardBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| VideoMME v2 (with subtitles) | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| VideoPhy | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| VideoScore-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ViDoRe v2 | Vision, audio and robotics | 3 of 3 | Nothing to fix | ||
| VidScribe | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ViewSpatial-Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| ViGiL3D | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| ViLLM-Eval | Knowledge and language | 8 of 8 | Nothing to fix | ||
| VIMA-Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Vinoground | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Virology Capabilities Test (VCT) | Safety and security | 11 of 11 | Nothing to fix | ||
| Virtual KITTI 2 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| VisFactor | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Vision2Web | Coding | 11 of 11 | Nothing to fix | ||
| Vision2Web S1-2026 | Coding | 9 of 9 | Nothing to fix | ||
| Vision2Web S2-2026 | Coding | 9 of 9 | Nothing to fix | ||
| Vision2Web S3-2026 | Coding | 10 of 10 | Nothing to fix | ||
| VisR-RAG-Fin | Professional work | 8 of 8 | Nothing to fix | ||
| Vistaar | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Visual Reasoning Benchmark (Fab AI) | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| VisualWebBench | Agents and tools | 8 of 8 | Nothing to fix | ||
| VisuLogic | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| VITA-Bench | Agents and tools | 11 of 11 | Nothing to fix | ||
| VitaBench 2.0 | Agents and tools | 10 of 10 | Nothing to fix | ||
| VitaBench-Noise (LongCat) | Agents and tools | 4 of 4 | Nothing to fix | ||
| VLADBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| VLMs Are Blind | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| VLMsAreBiased | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| VocalSound | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Voice Arena Arabic MSA | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Voice Arena Bahasa | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Voice Arena Brazilian Portuguese | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Voice Arena French | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Voice Arena Full Duplex Bench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Voice Arena German | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Voice Arena Hindi | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Voice Arena Indian English | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Voice Arena Japanese | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Voice Arena Mexican Spanish | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Voice Arena Tagalog | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Voice Arena US English | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Voice Arena Vietnamese | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| Voice Controllability Leaderboard | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Voice persuasion study on party preferences | Safety and security | 4 of 4 | Nothing to fix | ||
| Voice Replication Leaderboard | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| VoiceBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| VoiceCodeBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| VoiceMOS Challenge 2023 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| VoiceMOS Challenge 2024 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| VoiceMOS Challenge 2026 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| VoxCeleb1 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| VoxelBench Image | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| VoxelCodeBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| VoxLingua107 | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| VoxMem | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| VoxParity | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| VoxPopuli | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| VQAv2 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| VSI-Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| VSI-SUPER | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| VSR Random Split | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| VSR Zero-Shot Split | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| VSTI-Bench | Vision, audio and robotics | 5 of 5 | Nothing to fix | ||
| VTR-Bench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| WANDR | Agents and tools | 11 of 11 | Nothing to fix | ||
| Washington Post midterm voter guides | Knowledge and language | 8 of 8 | Nothing to fix | ||
| We-Math | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| WeaveBench | Agents and tools | 10 of 10 | Nothing to fix | ||
| Web Clone Bench | Coding | 11 of 11 | Nothing to fix | ||
| Web CTF challenges (Anthropic) | Safety and security | 4 of 4 | Nothing to fix | ||
| WebArena | Agents and tools | 11 of 11 | Nothing to fix | ||
| WebArena-Verified | Agents and tools | 6 of 6 | Nothing to fix | ||
| WebHarbor | Agents and tools | 9 of 9 | Nothing to fix | ||
| WebQuestions | Knowledge and language | 4 of 4 | Nothing to fix | ||
| WebShop | Agents and tools | 7 of 7 | Nothing to fix | ||
| WebsiteBench | Coding | 10 of 10 | Nothing to fix | ||
| WebSRC | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| WebTailBench-Refusals | Safety and security | 10 of 10 | Nothing to fix | ||
| WebVoyager | Agents and tools | 9 of 9 | Nothing to fix | ||
| WenetSpeech | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| WenetSpeech Test_Meeting | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| WenetSpeech Test_Net | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| WenetSpeech-Chuan-Eval | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Wer wird Millionär? LLM Benchmark | Knowledge and language | 9 of 9 | Nothing to fix | ||
| WHAM! | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| When2Call | Agents and tools | 10 of 10 | Nothing to fix | ||
| Which AI Model Knows Israel Best? | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Who Are You? LLM Self-Identity | Knowledge and language | 10 of 10 | Nothing to fix | ||
| WideSearch | Knowledge and language | 10 of 10 | Nothing to fix | ||
| Widget Captioning | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| WikiCompany2IndustryClustering | Professional work | 8 of 8 | Nothing to fix | ||
| WikiLingua | Knowledge and language | 8 of 8 | Nothing to fix | ||
| WikiSQL | Coding | 8 of 8 | Nothing to fix | ||
| WikiTableQuestions | Knowledge and language | 7 of 7 | Nothing to fix | ||
| WikiText-2 | Other | 8 of 8 | Nothing to fix | ||
| Wild-OmniDocBench | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| WildChat (toxic conversations) | Safety and security | 9 of 9 | Nothing to fix | ||
| WildChat non-toxic incorrect refusals (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| WildChat toxic refusals (Anthropic) | Safety and security | 3 of 3 | Nothing to fix | ||
| WildClawBench | Agents and tools | 11 of 11 | Nothing to fix | ||
| WildGraphBench | Knowledge and language | 8 of 8 | Nothing to fix | ||
| WildJailbreak | Safety and security | 7 of 7 | Nothing to fix | ||
| WildlifeBench Wildlife Knowledge V1 | Math and science | 10 of 10 | Nothing to fix | ||
| WildSpeech-Bench | Vision, audio and robotics | 9 of 9 | Nothing to fix | ||
| Winograd | Knowledge and language | 3 of 3 | Nothing to fix | ||
| WinoGrande | Knowledge and language | 9 of 9 | Nothing to fix | ||
| Winoground | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| WISE | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| WISER-Atomic | Agents and tools | 9 of 9 | Nothing to fix | ||
| WISER-Search | Agents and tools | 8 of 8 | Nothing to fix | ||
| Wisesight Sentiment | Knowledge and language | 7 of 7 | Nothing to fix | ||
| WitnessBench | Agents and tools | 11 of 11 | Nothing to fix | ||
| WMArena Video Arena | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| WMDP | Safety and security | 5 of 5 | Nothing to fix | ||
| WMDP Biology | Safety and security | 8 of 8 | Nothing to fix | ||
| WMDP Chemistry | Safety and security | 9 of 9 | Nothing to fix | ||
| WMDP Cyber | Safety and security | 8 of 8 | Nothing to fix | ||
| WMT 2023 Discourse-Level Literary Translation | Knowledge and language | 8 of 8 | Nothing to fix | ||
| WMT 2024 Discourse-Level Literary Translation | Knowledge and language | 8 of 8 | Nothing to fix | ||
| WMT23 | Knowledge and language | 3 of 3 | Nothing to fix | ||
| WMT24++ | Knowledge and language | 5 of 5 | Nothing to fix | ||
| WMT25 | Knowledge and language | 4 of 4 | Nothing to fix | ||
| WMT25 Terminology Translation Task | Knowledge and language | 4 of 4 | Nothing to fix | ||
| WMT26 | Knowledge and language | 4 of 4 | Nothing to fix | ||
| WMT26 Video Subtitle Translation | Knowledge and language | 7 of 7 | Nothing to fix | ||
| Wojak-Bench 3D | Professional work | 10 of 10 | Nothing to fix | ||
| WordEditBench | Professional work | 9 of 9 | Nothing to fix | ||
| WordleArenas | Agents and tools | 8 of 8 | Nothing to fix | ||
| Work360 | Professional work | 4 of 4 | Nothing to fix | ||
| WorkArena | Agents and tools | 7 of 7 | Nothing to fix | ||
| WorkBuddy User Blind Evaluation | Professional work | 8 of 8 | Nothing to fix | ||
| WorkBuddyBench-Office | Professional work | 9 of 9 | Nothing to fix | ||
| WorkflowArena | Agents and tools | 11 of 11 | Nothing to fix | ||
| WorkplaceAgents | Professional work | 11 of 11 | Nothing to fix | ||
| WorkRB Skill Extraction Challenge 2026 | Professional work | 8 of 8 | Nothing to fix | ||
| World-in-World | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| WorldArena | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| WorldBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| WorldExam | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| WorldModelBench | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| WorldOlympiad | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| WorldScore | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| WorldSense (omnimodal) | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| WorldVQA | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| WRENCH (Docker Compose) | Agents and tools | 8 of 8 | Nothing to fix | ||
| WRENCH (Factorio) | Agents and tools | 8 of 8 | Nothing to fix | ||
| Writing Benchmark (Every) | Knowledge and language | 8 of 8 | Nothing to fix | ||
| WritingBench | Knowledge and language | 10 of 10 | Nothing to fix | ||
| WSC | Knowledge and language | 8 of 8 | Nothing to fix | ||
| x-stance | Knowledge and language | 7 of 7 | Nothing to fix | ||
| XACLE Challenge 2026 | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Xbench | Other | 10 of 10 | Nothing to fix | ||
| xbench-DeepSearch | Agents and tools | 9 of 9 | Nothing to fix | ||
| xbench-ScienceQA | Math and science | 10 of 10 | Nothing to fix | ||
| XCOPA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| xDailyBench | Knowledge and language | 11 of 11 | Nothing to fix | ||
| xdev-bench-hard | Safety and security | 10 of 10 | Nothing to fix | ||
| XLRS-Bench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| XNLIeu | Knowledge and language | 8 of 8 | Nothing to fix | ||
| XOR-TyDi QA | Knowledge and language | 8 of 8 | Nothing to fix | ||
| XpertBench | Professional work | 4 of 4 | Nothing to fix | ||
| XSum | Knowledge and language | 8 of 8 | Nothing to fix | ||
| XTREME-R | Knowledge and language | 7 of 7 | Nothing to fix | ||
| XWinograd | Knowledge and language | 7 of 7 | Nothing to fix | ||
| YorùLect | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| YouCook2 Dense Video Description | Vision, audio and robotics | 7 of 7 | Nothing to fix | ||
| Your LLM doesn't Know BOUNDARY!. | Agents and tools | 10 of 10 | Nothing to fix | ||
| Your Voice, Your Device, Your Language Challenge | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| YouTube-VIS | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| YouTube-VOS | Vision, audio and robotics | 8 of 8 | Nothing to fix | ||
| Z.ai Code Bench | Coding | 11 of 11 | Nothing to fix | ||
| ZebraLogic | Math and science | 8 of 8 | Nothing to fix | ||
| ZeroBench | Vision, audio and robotics | 10 of 10 | Nothing to fix | ||
| ZeroSCROLLS/QuALITY | Knowledge and language | 8 of 8 | Nothing to fix | ||
| ZeroVideo | Vision, audio and robotics | 4 of 4 | Nothing to fix | ||
| ZhoBLiMP | Knowledge and language | 8 of 8 | Nothing to fix | ||
| ZMultiDialBench | Knowledge and language | 3 of 3 | Nothing to fix | ||
| ZMultiTransBench | Knowledge and language | 3 of 3 | Nothing to fix | ||
| τ-bench 2.0 | Agents and tools | 10 of 10 | Nothing to fix | ||
| τ-Voice | Vision, audio and robotics | 11 of 11 | Nothing to fix | ||
| τ^τ-bench | Coding | 10 of 10 | Nothing to fix | ||
| τ³-bench Banking | Agents and tools | 10 of 10 | Nothing to fix | ||
| τc-Bench | Agents and tools | 8 of 8 | Nothing to fix |
How we check
Each check is one simple rule. A check that doesn't apply to a benchmark is a note, never a problem, so an in-house safety eval with no public board isn't failing for lacking one.
Catalog record
- Task access
- Passes when the record says whether the tasks are public or private.
- Confidence
- Passes when we checked the record against a primary source. Likely means some facts rest on secondary sources. Unverified means we haven’t checked it yet.
Scores
- Independent runs
- Passes when someone other than the model’s maker measured scores: the official leaderboard, Epoch AI, or another board Epoch AI copied. An in-house eval can only have its lab’s numbers.
- Models scored
- Passes with five or more models, enough to rank them.
- Newest score
- Passes when a model got a score in the last six months. A test that saturated or was replaced can go quiet.
- Frontier models
- Passes when at least three of the ten strongest models on Epoch AI’s Capabilities Index have a score here. A retired test or an in-house eval doesn’t need them.
- Headroom
- Passes while the best score on a percent test is under 85%. Past that, we count it as saturated, which is a note, not a problem.
Leaderboard import
- Importer
- Passes when we read the official leaderboard ourselves. A test with no public leaderboard doesn’t need one.
- Last read
- Passes when we read the board in the week before the data’s date.
- Read worked
- Fails when the last attempt to read the board failed. Pages keep the rows from the last good read.
Consistency
- Lab claims
- Warns when a lab’s number for its own model is more than 10 points (or 10%) from an independent run of the same model, and we have no note on why.
- Known issues
- Fails while a published audit’s finding is still open. Disputed findings and fixes made only in a copy get a warning.
Problems that published audits found in a benchmark itself are on Audits.