31 labs report AIME 2024, and we have no independent scores for it yet
Finding its leaderboard is one of 1,237 open tasks below, from a missing release date to a leaderboard we don't read yet. Each links to a GitHub form that's already filled in. You add the value and a source, and a maintainer checks it.
The forms need a free GitHub account. Everything you write there is public.
How contributing works
- Pick a task and open its form. The form already names the record and what's missing. Add the value and a link to a source that shows it.
- A maintainer checks the source. When it holds up, they add the "accepted" label, and a bot opens a pull request with the change, linked to your issue.
- Code comes as pull requests. New importers and importer fixes go straight to a pull request. The steps are below.
You can also edit the data yourself. Every record lives in data/*.json, so a pull request that changes one record and names its source works as well as a form. To add a benchmark we don't have, use the submission form. The Submit page lists the other forms.
159 public benchmarks that two or more labs report have no independent scores here yet
Every score we have for these comes from the labs themselves. If a leaderboard or an outside group ranks models on one, send the link. If you made one of them, tell us where to read your results.
- AIME 2024 31 labs report it. Every score here is a lab's own.
- IFBench 24 labs report it. Every score here is a lab's own.
- BrowseComp 23 labs report it. Every score here is a lab's own.
- MATH 22 labs report it. Every score here is a lab's own.
- DROP 19 labs report it. Every score here is a lab's own.
Show 20 moreShow fewer
- MGSM 15 labs report it. Every score here is a lab's own.
- ChartQA 14 labs report it. Every score here is a lab's own.
- MMMLU 14 labs report it. Every score here is a lab's own.
- IMO-AnswerBench 13 labs report it. Every score here is a lab's own.
- TriviaQA 13 labs report it. Every score here is a lab's own.
- GDPval-AA v2 12 labs report it. Every score here is a lab's own.
- RefCOCO / RefCOCO+ / RefCOCOg 12 labs report it. Every score here is a lab's own.
- GDPval-AA 10 labs report it. Every score here is a lab's own.
- RealWorldQA 10 labs report it. Every score here is a lab's own.
- tau-bench 10 labs report it. Every score here is a lab's own.
- FLEURS 9 labs report it. Every score here is a lab's own.
- ScreenSpot-v2 9 labs report it. Every score here is a lab's own.
- GraphWalks 8 labs report it. Every score here is a lab's own.
- LibriSpeech 8 labs report it. Every score here is a lab's own.
- LogicVista 8 labs report it. Every score here is a lab's own.
- MRCR v2 8 labs report it. Every score here is a lab's own.
- XSTest 8 labs report it. Every score here is a lab's own.
- BBQ (Bias Benchmark for QA) 7 labs report it. Every score here is a lab's own.
- CVBench 7 labs report it. Every score here is a lab's own.
- GDPval-AA v2.1 7 labs report it. Every score here is a lab's own.
The other 134 are in tasks.json, with the rest of this page.
287 benchmarks have no release date
The month its paper, repository or launch post first appeared. Give the link you dated it from.
- Apollo Research strategic deception evaluations No release date.
- GRE Quantitative No release date.
- GRE Verbal No release date.
- LSAT No release date.
- MMBench-CN v1.1 No release date.
Show 20 moreShow fewer
- Molecular Biology Capabilities Test No release date.
- AP Art History No release date.
- AP Biology No release date.
- AP Chemistry No release date.
- AP English Language and Composition No release date.
- AP English Literature and Composition No release date.
- AP Environmental Science No release date.
- AP Macroeconomics No release date.
- AP Microeconomics No release date.
- AP Psychology No release date.
- AP Statistics No release date.
- AP US Government No release date.
- AP US History No release date.
- AP World History No release date.
- GRE Analytical Writing No release date.
- Harvey LAB-AA v1.0 No release date.
- SAT Math No release date.
- World-Class Bio No release date.
- Advanced Sommelier (theory knowledge) No release date.
- Agent behavior under-service and overreach eval No release date.
The other 262 are in tasks.json, with the rest of this page.
26 benchmarks have no difficulty level
Who could solve its tasks, from most adults to researchers on open problems. The levels are explained on the About page. See the levels.
- LongFact No difficulty level.
- ASIMOV No difficulty level.
- Autointent-Bio No difficulty level.
- Autointent-Chem No difficulty level.
- CyberSecEval Spear Phishing No difficulty level.
Show 20 moreShow fewer
- HealthBench Hard Hallucinations No difficulty level.
- MIEB Lite No difficulty level.
- OpenAI performance in cases flagged by users No difficulty level.
- OpenAI user-flagged hallucination eval No difficulty level.
- PAI-Bench No difficulty level.
- Physics-IQ No difficulty level.
- QwenWorldBench No difficulty level.
- RBench No difficulty level.
- SimpleQA Hallucination Rate No difficulty level.
- VideoPhy-2 No difficulty level.
- xAI Factuality (Hallucination) eval No difficulty level.
- 1X World Model Challenge No difficulty level.
- BackSearch No difficulty level.
- DreamGen Bench No difficulty level.
- EWMBench No difficulty level.
- WorldArena No difficulty level.
- WorldArena 2.0 No difficulty level.
- WorldModelBench No difficulty level.
- WorldModelGym No difficulty level.
- WorldScore No difficulty level.
The other 1 are in tasks.json, with the rest of this page.
474 public benchmarks have no paper or report linked
The paper, technical report or launch post that introduced it. Some have none, and that is worth knowing too.
- Berkeley Function Calling Leaderboard (BFCL) No paper or report linked.
- Terminal-Bench 4.0 No paper or report linked.
- MMMLU No paper or report linked.
- τ³-bench Banking No paper or report linked.
- AA-LCR (Artificial Analysis Long Context Reasoning) No paper or report linked.
Show 20 moreShow fewer
- BFCL V4 No paper or report linked.
- Toolathlon-Verified No paper or report linked.
- GDPval-AA No paper or report linked.
- RealWorldQA No paper or report linked.
- Aider Polyglot No paper or report linked.
- GraphWalks No paper or report linked.
- LibriSpeech No paper or report linked.
- PinchBench No paper or report linked.
- Terminal-Bench Hard No paper or report linked.
- AA-LCR 1.1 No paper or report linked.
- Needle In A Haystack No paper or report linked.
- OpenAI MRCR No paper or report linked.
- OpenAI MRCR v2 No paper or report linked.
- Terminal-Bench 3.0 No paper or report linked.
- Terminal-Bench-Science No paper or report linked.
- BioMysteryBench No paper or report linked.
- Creative Writing v3 No paper or report linked.
- Big Bench Audio No paper or report linked.
- Aider-Edit No paper or report linked.
- EQ-Bench 3 No paper or report linked.
The other 449 are in tasks.json, with the rest of this page.
158 models have no release date
The day its lab released it, from the launch post, model card or API changelog.
- BAGEL No release date.
- MiniMax H3 No release date.
- GPT-Image-1.5 No release date.
- GLM-Image No release date.
- GPT-Image-2 No release date.
Show 20 moreShow fewer
- FLUX.2 [dev] No release date.
- Seedream 4.0 No release date.
- FLUX.2 [klein] 4B No release date.
- GPT-Image-1 No release date.
- FLUX.2 [klein] 9B No release date.
- HappyHorse 1.0 No release date.
- Qwen-Image No release date.
- FLUX.2 [pro] No release date.
- Gemini Omni Flash No release date.
- Grok Imagine Image 2.0 No release date.
- HunyuanImage 3.0 No release date.
- Seedream 5.0 Pro No release date.
- FLUX.1 [dev] No release date.
- Gemini 2.5 Flash Image Preview (nano-banana) No release date.
- Gemini Nano Banana 2.1 No release date.
- Gemini Omni 1.1 Flash No release date.
- GPT-Image-2.5 Flare No release date.
- GPT-Image-2.5 Sunburst No release date.
- Qwen-Image (2512) No release date.
- Qwen-Image-3.0-Pro No release date.
The other 133 are in tasks.json, with the rest of this page.
41 new benchmarks wait for review
Our daily scan ranked these worth cataloging. A benchmark gets in when it passes one of four tests, so evidence that a lab reports it or a leaderboard runs it settles it.
- BestRand Random Is Almost Enough? Stable and Low-Cost Benchmark Compression via BestRand
- Co-Design Bench Co-Design Bench: Multimodal Collaborative Coding Evaluation for Agentic Models
- CrossSWE CrossSWE: Benchmarking Cross-Repository Joint Development for Coding Agents
- FinDendro LLMs Don't Know What They Know: Benchmarking Temporal Self-Knowledge for Financial Backtesting
- Infrathon Infrathon: Benchmarking Language Agents for Real-World, Long-Horizon AI Infra Optimization
Show 20 moreShow fewer
- Kierston & KierstonBench Kierston & KierstonBench: Simulation of Allentown, PA and Evaluation of Language Models in Crime Investigation and Surveillance Tasks
- LongBanana LongBanana: An Expert-Verified Benchmark for Long-Context Multi-Reference Image Synthesis
- TB-MathModel Terminal-style Mathematical Modeling Bench: The First Agentic and Verifiable Benchmark for Mathematical Modeling
- τ-Rec-Books $τ$-Rec-Books: A Disclosure-Controlled Benchmark for Requirement Elicitation in Agentic Recommender Systems
- Beyond Instruction Override Beyond Instruction Override: Benchmarking Indirect Prompt Injection with Task-Tampering Attacks
- DRPixelBench DR Pixel: Page-Native Deep Research with Vision-Language Models
- EarthRecord EarthRecord: From Pixels to Land Change with Multimodal Agents over Four Decades of Satellite Imagery
- Embodied Agent Arena Are Frontier VLM Agents Ready to Be Robot Generalists? An Empirical Study with the Embodied Agent Arena
- GeoFaham-Bench GeoFaham: Domain-Specialized Multi-Agent Geospatial Analytics for Disaster Response
- IdeaTaste IdeaTaste: Do LLMs Share the Research Taste of Human Researchers?
- Longtail-SWE Longtail-SWE: High-Quality SWE Tasks for Underrepresented Languages
- Lost in Recovery Lost in Recovery: Demystifying Recovery Behavior of Computer-Use Agents via Goal-Oriented Action Replay
- MMDM MMDM: Measuring Massive Multitask Decision Making
- Multimodal Knowledge Agent Introducing Gemini 4 Hydrogen
- One Table Benchmark One Table, Many Queries: Benchmarking Generalist Tabular Models Across Generation, Imputation, and Prediction
- RealSSA Efficient Streaming Audio-Visual Target Speaker Extraction for Real-World Acoustic Scenes
- RoboColosseum RoboColosseum: Evaluating Physical Intelligence at Scale In-The-Wild
- VidUEU-Agent VidUEU-Agent: A Data-Curation Agent for Multimodal Understanding, Editing, and Unified Tasks
- WebsiteBench Can AI agents rebuild websites through browser exploration? 70 website-reconstruction tasks.
- ASIMOV-Jailbreak Gemini Robotics 2.5 accelerates real-world general dexterity
The other 16 are in tasks.json, with the rest of this page.
5 write-ups may criticize a benchmark we catalog
Our scan flagged these. If one documents a real problem, such as leaked answers or a broken grader, log it as a critique.
- HANS Can the Confidence and Decisions of System 1 Models Be Trusted? An Explainability and Perturbation Audit of Jev-Style Decision Models
- Internal agentic coding evaluation (Anthropic) Construct Validity of Coding Benchmarks for LLM-based Agentic Coding Tools
- LegalBench, Legora BAR Follow the Policy, Fail the Benchmark: Auditing and Revising Policy-Based Decision Benchmarks
- MedQA, MedMCQA, PubMedQA From Small to Large: Assessing Medical Reasoning, Benchmark Contamination, and Model Size in Language Models
- MIMIC-CXR (classification) Leakage- and Label-Aware Benchmarking of Class-Imbalance Strategies for Multimodal Chest X-Ray and EHR Classification
79 official leaderboards we link aren't read yet
Each needs a short importer that reads the board's published data. Check the board's terms first. How to write one.
- Common Voice 6 labs report it. We link its leaderboard but don't read it.
- CyScenarioBench 5 labs report it. We link its leaderboard but don't read it.
- Natural Questions 5 labs report it. We link its leaderboard but don't read it.
- EvalPlus 4 labs report it. We link its leaderboard but don't read it.
- OmniDocBench 4 labs report it. We link its leaderboard but don't read it.
Show 20 moreShow fewer
- Global MMLU Lite 3.0 3 labs report it. We link its leaderboard but don't read it.
- MASK 3 labs report it. We link its leaderboard but don't read it.
- MVPBench 3 labs report it. We link its leaderboard but don't read it.
- WorkSpaceBench 3 labs report it. We link its leaderboard but don't read it.
- BankerToolBench 2 labs report it. We link its leaderboard but don't read it.
- CursorBench 3.1 2 labs report it. We link its leaderboard but don't read it.
- CursorBench 3.2 2 labs report it. We link its leaderboard but don't read it.
- CyberGym (final-submission metric) 1 lab reports it. We link its leaderboard but don't read it.
- FORTRESS-RN 1 lab reports it. We link its leaderboard but don't read it.
- Global MMLU Lite 2.0 1 lab reports it. We link its leaderboard but don't read it.
- Jigsaw Toxic Comment Classification 1 lab reports it. We link its leaderboard but don't read it.
- KernelBench 1 lab reports it. We link its leaderboard but don't read it.
- LVBench (with memory) 1 lab reports it. We link its leaderboard but don't read it.
- Matbench 1 lab reports it. We link its leaderboard but don't read it.
- Meeseeks 1 lab reports it. We link its leaderboard but don't read it.
- PMPP-Hard 1 lab reports it. We link its leaderboard but don't read it.
- ProteinGym 1 lab reports it. We link its leaderboard but don't read it.
- RelBench 1 lab reports it. We link its leaderboard but don't read it.
- TydiQA 1 lab reports it. We link its leaderboard but don't read it.
- A.S.E (AICGSecEval) We link its leaderboard but don't read it.
The other 54 are in tasks.json, with the rest of this page.
8 importers have stopped working
The board keeps its last scores until the importer is fixed. Most breaks are a moved file or a changed page. How importers work.
- DocVQA It failed on 2026-11-12: Command failed: curl --fail --silent --show-error --max-time 40 --user-agent benchmark-almanac (+https://github.com/brianlzhou/benchmark-almanac) https://rrc.cvc.uab.es/?ch=17&com=evaluation&task=1 cu.
- DUDE It failed on 2026-11-12: Command failed: curl --fail --silent --show-error --max-time 40 --user-agent benchmark-almanac (+https://github.com/brianlzhou/benchmark-almanac) https://rrc.cvc.uab.es/?ch=23&com=evaluation&task=1 cu.
- Cekura Production Voice Agents It failed on 2026-11-12: Cekura table missing.
- Cooperate to Compete (C2C) It failed on 2026-11-12: The operation was aborted due to timeout.
- Human Pathogen Capabilities Test v2 (HPCT v2) It failed on 2026-11-12: no chart matches /Model performance on HPCT-v2$/.
Show 3 moreShow fewer
- Leipzig Benchmark It failed on 2026-11-12: Leipzig model scores table not found.
- StrataBench Land v1 Extract It failed on 2026-11-12: 401 https://huggingface.co/datasets/Collide-ai/stratabench-land-v1/raw/main/README.md.
- VideoFDB It failed on 2026-11-12: No videofdb-nvidia score rows.
Write an importer
Each file in scripts/leaderboards/ reads one benchmark's official leaderboard, and 1,746 boards already have one. Most are short, because shared readers handle common sources such as Hugging Face datasets and GitHub tables.
- Pick a board from the leaderboards above, and check that its terms and robots.txt allow reading it.
- Start from a short importer, such as
scripts/leaderboards/aider-polyglot.ts, saved asscripts/leaderboards/<benchmark-id>.ts. Point it at the board's published data file. - Test it with
node scripts/sync-leaderboards.ts --only <benchmark-id> --dry, and open a pull request.
The importer guide explains every field and the rules for each source.