Benchmark Almanac

Of the 901 benchmarks labs named in 2026, 585 were named by just one lab

We counted every benchmark named in the 266 launch posts labs published in 2026, and in the 139 system cards, model cards and technical reports published with them. Each shape below is one benchmark. The more labs named it, the further right it sits, and the more releases named it, the bigger it is.

The frontier labs are the nine whose best 2026 models score highest on Epoch AI's Capabilities Index, as of November 2026: Anthropic, OpenAI, Google DeepMind, Moonshot AI, Meta, DeepSeek, xAI, Qwen and Z.ai. A frontier model ranked in the top ten on Epoch AI's Capabilities Index among the models out the day it was released.

123451015202530Labs that named it in 2026585 of 901named by one labLiveCodeBench26SWE-Bench Pro28GPQA Diamond29tau-bench31AIME31Terminal-Bench33IFBench24SWE-bench Multilingual25Humanity's Last Exam27SWE-bench Verified31Labs that named it123451015202530585 of 901named by one labTerminal-Bench33SWE-bench Verified31tau-bench31AIME31GPQA Diamond29SWE-Bench Pro28Humanity's Last Exam27LiveCodeBench26
  • Knowledge and language
  • Math and science
  • Coding
  • Agents and tools
  • Professional work
  • Vision, audio and robotics
  • Safety and security
  • Other

releases named it

Source: Labs’ launch posts, read by Benchmark Almanac

How has reporting changed?

In the first half of 2024, 34% of the benchmarks labs named had come out in the year before the release. In the second half of 2026 so far, 63% had. The ten most-named benchmarks went from 35% of mentions to 20%.

Released in the past year

Named benchmarks that came out in the 12 months before the release

0% 20% 40% 60% 80% 2024 2025 2026 34% 63% 0% 20% 40% 60% 80% 2024 2025 2026 34% 63%

The ten most named

Mentions that went to the ten benchmarks named most that half year, with versions of a test as one

0% 10% 20% 30% 40% 2024 2025 2026 35% 20% 0% 10% 20% 30% 40% 2024 2025 2026 35% 20%

The lab’s own benchmarks

Named benchmarks that the releasing lab made

0% 5% 10% 15% 2024 2025 2026 13% 13% 0% 5% 10% 15% 2024 2025 2026 13% 13%

Benchmarks per release

The median release, counting each benchmark once

0 10 20 30 2024 2025 2026 21 13 0 10 20 30 2024 2025 2026 21 13

Source: Labs’ launch posts, read by Benchmark Almanac

How has reporting changed?

In the first half of 2024, 39% of the benchmarks labs named had come out in the year before the release. In the second half of 2026 so far, 65% had. The ten most-named benchmarks went from 30% of mentions to 14%.

Released in the past year

Named benchmarks that came out in the 12 months before the release

0% 20% 40% 60% 80% 2024 2025 2026 39% 65% 0% 20% 40% 60% 80% 2024 2025 2026 39% 65%

The ten most named

Mentions that went to the ten benchmarks named most that half year, with versions of a test as one

0% 10% 20% 30% 2024 2025 2026 30% 14% 0% 10% 20% 30% 2024 2025 2026 30% 14%

The lab’s own benchmarks

Named benchmarks that the releasing lab made

0% 10% 20% 30% 40% 2024 2025 2026 19% 31% 0% 10% 20% 30% 40% 2024 2025 2026 19% 31%

Benchmarks per release

The median release, counting each benchmark once

0 10 20 30 2024 2025 2026 25 16 0 10 20 30 2024 2025 2026 25 16

Source: Labs’ launch posts, read by Benchmark Almanac

How has reporting changed?

In the first half of 2024, 40% of the benchmarks labs named for frontier models had come out in the year before the release. In the second half of 2026 so far, 77% had. The ten most-named benchmarks went from 32% of mentions to 25%.

Released in the past year

Named benchmarks that came out in the 12 months before the release

0% 20% 40% 60% 80% 2024 2025 2026 40% 77% 0% 20% 40% 60% 80% 2024 2025 2026 40% 77%

The ten most named

Mentions that went to the ten benchmarks named most that half year, with versions of a test as one

0% 10% 20% 30% 40% 2024 2025 2026 32% 25% 0% 10% 20% 30% 40% 2024 2025 2026 32% 25%

The lab’s own benchmarks

Named benchmarks that the releasing lab made

0% 10% 20% 30% 2024 2025 2026 13% 25% 0% 10% 20% 30% 2024 2025 2026 13% 25%

Benchmarks per release

The median release, counting each benchmark once

0 10 20 30 40 2024 2025 2026 23 32 0 10 20 30 40 2024 2025 2026 23 32

Source: Labs’ launch posts, read by Benchmark Almanac

How has reporting changed?

In the first half of 2024, 46% of the benchmarks labs named for frontier models had come out in the year before the release. In the second half of 2026 so far, 76% had. The ten most-named benchmarks went from 28% of mentions to 11%.

Released in the past year

Named benchmarks that came out in the 12 months before the release

0% 20% 40% 60% 80% 2024 2025 2026 46% 76% 0% 20% 40% 60% 80% 2024 2025 2026 46% 76%

The ten most named

Mentions that went to the ten benchmarks named most that half year, with versions of a test as one

0% 10% 20% 30% 2024 2025 2026 28% 11% 0% 10% 20% 30% 2024 2025 2026 28% 11%

The lab’s own benchmarks

Named benchmarks that the releasing lab made

0% 20% 40% 60% 2024 2025 2026 20% 55% 0% 20% 40% 60% 2024 2025 2026 20% 55%

Benchmarks per release

The median release, counting each benchmark once

0 20 40 60 80 2024 2025 2026 23 62 0 20 40 60 80 2024 2025 2026 23 62

Source: Labs’ launch posts, read by Benchmark Almanac

How has reporting changed?

In the first half of 2024, 37% of the benchmarks frontier labs named had come out in the year before the release. In the second half of 2026 so far, 73% had. The ten most-named benchmarks went from 34% of mentions to 23%.

Released in the past year

Named benchmarks that came out in the 12 months before the release

0% 20% 40% 60% 80% 2024 2025 2026 37% 73% 0% 20% 40% 60% 80% 2024 2025 2026 37% 73%

The ten most named

Mentions that went to the ten benchmarks named most that half year, with versions of a test as one

0% 10% 20% 30% 40% 2024 2025 2026 34% 23% 0% 10% 20% 30% 40% 2024 2025 2026 34% 23%

The lab’s own benchmarks

Named benchmarks that the releasing lab made

0% 5% 10% 15% 20% 2024 2025 2026 14% 19% 0% 5% 10% 15% 20% 2024 2025 2026 14% 19%

Benchmarks per release

The median release, counting each benchmark once

0 10 20 30 2024 2025 2026 21 17 0 10 20 30 2024 2025 2026 21 17

Source: Labs’ launch posts, read by Benchmark Almanac

How has reporting changed?

In the first half of 2024, 41% of the benchmarks frontier labs named had come out in the year before the release. In the second half of 2026 so far, 72% had. The ten most-named benchmarks went from 29% of mentions to 13%.

Released in the past year

Named benchmarks that came out in the 12 months before the release

0% 20% 40% 60% 80% 2024 2025 2026 41% 72% 0% 20% 40% 60% 80% 2024 2025 2026 41% 72%

The ten most named

Mentions that went to the ten benchmarks named most that half year, with versions of a test as one

0% 10% 20% 30% 2024 2025 2026 29% 13% 0% 10% 20% 30% 2024 2025 2026 29% 13%

The lab’s own benchmarks

Named benchmarks that the releasing lab made

0% 20% 40% 60% 2024 2025 2026 20% 43% 0% 20% 40% 60% 2024 2025 2026 20% 43%

Benchmarks per release

The median release, counting each benchmark once

0 10 20 30 2024 2025 2026 25 23 0 10 20 30 2024 2025 2026 25 23

Source: Labs’ launch posts, read by Benchmark Almanac

How has reporting changed?

In the first half of 2024, 42% of the benchmarks frontier labs named for frontier models had come out in the year before the release. In the second half of 2026 so far, 77% had. The ten most-named benchmarks went from 31% of mentions to 25%.

Released in the past year

Named benchmarks that came out in the 12 months before the release

0% 20% 40% 60% 80% 2024 2025 2026 42% 77% 0% 20% 40% 60% 80% 2024 2025 2026 42% 77%

The ten most named

Mentions that went to the ten benchmarks named most that half year, with versions of a test as one

0% 10% 20% 30% 40% 2024 2025 2026 31% 25% 0% 10% 20% 30% 40% 2024 2025 2026 31% 25%

The lab’s own benchmarks

Named benchmarks that the releasing lab made

0% 10% 20% 30% 2024 2025 2026 13% 25% 0% 10% 20% 30% 2024 2025 2026 13% 25%

Benchmarks per release

The median release, counting each benchmark once

0 10 20 30 40 2024 2025 2026 28 32 0 10 20 30 40 2024 2025 2026 28 32

Source: Labs’ launch posts, read by Benchmark Almanac

How has reporting changed?

In the first half of 2024, 48% of the benchmarks frontier labs named for frontier models had come out in the year before the release. In the second half of 2026 so far, 76% had. The ten most-named benchmarks went from 27% of mentions to 11%.

Released in the past year

Named benchmarks that came out in the 12 months before the release

0% 20% 40% 60% 80% 2024 2025 2026 48% 76% 0% 20% 40% 60% 80% 2024 2025 2026 48% 76%

The ten most named

Mentions that went to the ten benchmarks named most that half year, with versions of a test as one

0% 10% 20% 30% 2024 2025 2026 27% 11% 0% 10% 20% 30% 2024 2025 2026 27% 11%

The lab’s own benchmarks

Named benchmarks that the releasing lab made

0% 20% 40% 60% 2024 2025 2026 20% 55% 0% 20% 40% 60% 2024 2025 2026 20% 55%

Benchmarks per release

The median release, counting each benchmark once

0 20 40 60 80 2024 2025 2026 29 62 0 20 40 60 80 2024 2025 2026 29 62

Source: Labs’ launch posts, read by Benchmark Almanac

Do lab numbers match independent runs?

In 2026 so far, labs' own numbers beat an independent run of the same model on the same benchmark in 56% of the pairs where the two differ, by a median of 0.7 points. In 2024 it was 64%. This counts every lab, whatever the switches above say.

Every pair

The lab’s number is higher

Of the pairs where the two differ

0% 25% 50% 75% 100% 2024 2025 2026 64% 56% 0% 25% 50% 75% 100% 2024 2025 2026 64% 56%

Median gap, in points

The lab’s number less the independent run’s

0 0.5 1 1.5 2 2024 2025 2026 0.9 0.7 0 0.5 1 1.5 2 2024 2025 2026 0.9 0.7

Sources: Official leaderboards, Epoch AI (CC BY 4.0) and labs’ model cards

Which benchmarks stay on top?

We track the ten benchmarks labs cite in the most releases each quarter. Only GPQA Diamond has made the top ten in every quarter since 2024. Terminal-Bench has been first for three quarters running.

2024 whole year Q1 2025 Q2 2025 Q3 2025 Q4 2025 Q1 2026 Q2 2026 Q3 2026 Q4 2026 so far 1st 10th MMLU HumanEval MATH GPQA Diamond GPQA Diamond GSM8K BIG-Bench Hard MGSM MBPP DROP ARC-Challenge Humanity's Last Exam Terminal-Bench DeepSWE AutomationBench OSWorld Agents' Last Exam Terminal-Bench-Science FrontierCode Chartography MMLU out HumanEval out MATH out LiveCodeBench LiveCodeBench out MMLU-Pro MMLU-Pro out IFEval IFEval out AIME AIME out MMMU MMMU out SimpleQA SimpleQA SWE-bench Verified SWE-bench Verified out Aider Humanity's Last Exam tau-bench tau-bench out Terminal-Bench MMMLU IFBench MMMU-Pro SWE-Bench Pro SWE-Bench Pro out DeepSWE GDPval-AA Toolathlon OSWorld Agents' Last Exam FrontierCode

Source: Labs’ launch posts, read by Benchmark Almanac

Which benchmarks stay on top?

We track the ten benchmarks labs cite in the most frontier-model releases each quarter. GPQA Diamond and SWE-bench Verified each made the top ten in six of the eight quarters since 2025 began, more than any other benchmark.

2024 whole year Q1 2025 Q2 2025 Q3 2025 Q4 2025 Q1 2026 Q2 2026 Q3 2026 Q4 2026 so far 1st 10th MMLU HumanEval MATH GPQA Diamond MGSM DROP GSM8K BIG-Bench Hard ARC-Challenge MBPP Terminal-Bench DeepSWE AutomationBench HealthBench Professional BenchCAD Agents' Last Exam Terminal-Bench-Science Chartography FrontierSWE Artificial Analysis Intelli… GPQA Diamond out AIME AIME out LiveCodeBench LiveCodeBench MMMU MMMU out SimpleQA SimpleQA out MMLU-Pro SWE-bench Verified SWE-bench Verified out MATH-500 EgoSchema IFEval Aider Aider out Codeforces rating Humanity's Last Exam out MMMLU MMMLU tau-bench tau-bench out HMMT HMMT out Terminal-Bench Terminal-Bench out Terminal-Bench ARC-AGI ARC-AGI out BrowseComp BrowseComp out MMMU-Pro GDPval-AA GDPval-AA out MCP Atlas MCP Atlas out OSWorld OSWorld out SWE-Bench Pro CharXiv DeepSWE HealthBench Professional FrontierCode Toolathlon BenchCAD Terminal-Bench-Science Chartography FrontierSWE

Source: Labs’ launch posts, read by Benchmark Almanac

Which benchmarks stay on top?

We track the ten benchmarks frontier labs cite in the most releases each quarter. Humanity's Last Exam made frontier labs' top ten in seven of the eight quarters since 2025 began, more than any other benchmark.

2024 whole year Q1 2025 Q2 2025 Q3 2025 Q4 2025 Q1 2026 Q2 2026 Q3 2026 Q4 2026 so far 1st 10th MMLU HumanEval MATH GPQA Diamond GSM8K BIG-Bench Hard MGSM DROP MMMU MBPP Humanity's Last Exam Terminal-Bench DeepSWE AutomationBench Agents' Last Exam HealthBench Professional Terminal-Bench-Science Chartography Artificial Analysis Intelli… ProgramBench MMLU out MATH out GPQA Diamond out MMMU out LiveCodeBench LiveCodeBench out LiveCodeBench LiveCodeBench out MMLU-Pro AIME AIME out AIME SimpleQA SimpleQA out IFEval EgoSchema SWE-bench Verified SWE-bench Verified out SWE-bench Verified Aider Aider out MMMLU MMMLU MathVista Humanity's Last Exam tau-bench tau-bench out HMMT HMMT out Terminal-Bench ARC-AGI BrowseComp MMMU-Pro MMMU-Pro out OmniDocBench CharXiv CharXiv out OSWorld OSWorld SWE-Bench Pro SWE-Bench Pro out DeepSWE AutomationBench GDPval-AA Toolathlon HealthBench Professional Terminal-Bench-Science Chartography

Source: Labs’ launch posts, read by Benchmark Almanac

Which benchmarks stay on top?

We track the ten benchmarks frontier labs cite in the most frontier-model releases each quarter. GPQA Diamond and SWE-bench Verified each made frontier labs' top ten in six of the eight quarters since 2025 began, more than any other benchmark.

2024 whole year Q1 2025 Q2 2025 Q3 2025 Q4 2025 Q1 2026 Q2 2026 Q3 2026 Q4 2026 so far 1st 10th MMLU GPQA Diamond HumanEval MATH MGSM BIG-Bench Hard DROP GSM8K MathVista MMMU Terminal-Bench DeepSWE AutomationBench HealthBench Professional BenchCAD Agents' Last Exam Terminal-Bench-Science Chartography FrontierSWE Artificial Analysis Intelli… GPQA Diamond out MMMU out AIME AIME out LiveCodeBench LiveCodeBench SimpleQA SimpleQA out MMLU-Pro SWE-bench Verified SWE-bench Verified out MATH-500 EgoSchema IFEval Aider Aider out Codeforces rating Humanity's Last Exam out MMMLU MMMLU tau-bench tau-bench out HMMT HMMT out Terminal-Bench Terminal-Bench out Terminal-Bench ARC-AGI ARC-AGI out BrowseComp BrowseComp out MMMU-Pro GDPval-AA GDPval-AA out MCP Atlas MCP Atlas out OSWorld OSWorld out SWE-Bench Pro CharXiv DeepSWE HealthBench Professional FrontierCode Toolathlon BenchCAD Terminal-Bench-Science Chartography FrontierSWE

Source: Labs’ launch posts, read by Benchmark Almanac

How fast do new benchmarks catch on?

DeepSWE reached ten labs 38 days after the first one cited it, the fastest of the 29 new benchmarks that got there. Half of them took 211 days or less. We count every benchmark released since 2025 that at least five labs have cited.

0 10 20 30 40 0 3 6 9 12 15 18 21 24 Months since a lab first named it Terminal-Bench Humanity's Last Exam DeepSWE, ten labs in 38 days τ³-bench Banking IMO-AnswerBench Agents' Last Exam HMMT Nov 2025 0 10 20 30 40 0 6 12 18 24 Months since a lab first named it Terminal-Bench Humanity's Last Exam DeepSWE τ³-bench Banking IMO-AnswerBench Agents' Last Exam HMMT Nov 2025

Source: Labs’ launch posts, read by Benchmark Almanac

How fast do new benchmarks catch on?

Humanity's Last Exam is the only new benchmark that reached eight labs, 379 days after the first one cited it. We count every benchmark released since 2025 that at least three labs have cited for frontier models.

0 1 2 3 4 5 6 7 8 0 3 6 9 12 15 18 21 Months since a lab first named it Humanity's Last Exam, eight labs in 379 days 0 1 2 3 4 5 6 7 8 0 6 12 18 Months since a lab first named it Humanity's Last Exam

Source: Labs’ launch posts, read by Benchmark Almanac

How fast do new benchmarks catch on?

DeepSWE reached all nine frontier labs 48 days after the first one cited it, the fastest of the five new benchmarks that got there. Half of them took 234 days or less. We count every benchmark released since 2025 that at least five frontier labs have cited.

0 1 2 3 4 5 6 7 8 9 0 3 6 9 12 15 18 21 24 Months since a lab first named it DeepSWE, all nine in 48 days GDPval-AA SWE-Bench Pro Terminal-Bench Humanity's Last Exam 0 1 2 3 4 5 6 7 8 9 0 6 12 18 24 Months since a lab first named it Humanity's Last Exam Terminal-Bench SWE-Bench Pro GDPval-AA DeepSWE

Source: Labs’ launch posts, read by Benchmark Almanac

How fast do new benchmarks catch on?

Humanity's Last Exam is the only new benchmark that reached all eight frontier labs, 379 days after the first one cited it. We count every benchmark released since 2025 that at least three frontier labs have cited for frontier models.

0 1 2 3 4 5 6 7 8 0 3 6 9 12 15 18 21 Months since a lab first named it Humanity's Last Exam, all eight in 379 days 0 1 2 3 4 5 6 7 8 0 6 12 18 Months since a lab first named it Humanity's Last Exam

Source: Labs’ launch posts, read by Benchmark Almanac

Which benchmarks do one or two labs carry?

Of the 332 benchmarks named in five or more releases since 2024, 144 owe at least half of those releases to one lab, and 7 more owe four in five to two labs. 41 of the ones a single lab carries are its own. Qwen carries the most that others made, 50, among them C-Eval and SuperGPQA.

Which benchmarks do one or two labs carry?

Of the 491 benchmarks named in five or more releases since 2024, 292 owe at least half of those releases to one lab, and 8 more owe four in five to two labs. 164 of the ones a single lab carries are its own. Qwen carries the most that others made, 50, among them C-Eval and SuperGPQA.

  1. C-Eval Qwen 50% 36
  2. SuperGPQA Qwen 56% 36
  3. MMLU-Redux Qwen 71% 35
  4. RealWorldQA Qwen 57% 35
  5. ERQA Qwen 55% 33
  6. LVBench Qwen 53% 32
  7. RefCOCO / RefCOCO+ / RefCOCOg Qwen 50% 28
  8. INCLUDE Qwen 60% 25
  9. CountBench Qwen 58% 24
  10. AndroidWorld Qwen 61% 23
  11. DynaMath Qwen 52% 23
  12. MMLU-ProX Qwen 61% 23
  13. MLVU Qwen 77% 22
  14. MVBench Qwen 64% 22
  15. MedXpertQA-MM Qwen 52% 21
  16. LiveBench Qwen 55% 20
  17. MMLongBench-Doc Qwen 58% 19
  18. Global MMLU Lite Google DeepMind 67% 18
  19. V* Qwen 53% 17
  20. VITA-Bench Qwen 65% 17
  1. MRCR Google DeepMind 70% its own 37
  2. C-Eval Qwen 50% 36
  3. SuperGPQA Qwen 56% 36
  4. MMLU-Redux Qwen 71% 35
  5. RealWorldQA Qwen 57% 35
  6. ERQA Qwen 55% 33
  7. LVBench Qwen 53% 32
  8. OpenAI MRCR OpenAI 50% its own 28
  9. RefCOCO / RefCOCO+ / RefCOCOg Qwen 50% 28
  10. INCLUDE Qwen 60% 25
  11. CountBench Qwen 58% 24
  12. AndroidWorld Qwen 61% 23
  13. DynaMath Qwen 52% 23
  14. MMLU-ProX Qwen 61% 23
  15. MLVU Qwen 77% 22
  16. MVBench Qwen 64% 22
  17. HealthBench Hard OpenAI 76% its own 21
  18. MedXpertQA-MM Qwen 52% 21
  19. LiveBench Qwen 55% 20
  20. TroubleshootingBench OpenAI 100% its own 20

Source: Labs’ launch posts, read by Benchmark Almanac

  1. BBQ (Bias Benchmark for QA) Anthropic 55% 38
  2. C-Eval Qwen 50% 36
  3. SuperGPQA Qwen 56% 36
  4. MMLU-Redux Qwen 71% 35
  5. RealWorldQA Qwen 57% 35
  6. Virology Capabilities Test (VCT) Anthropic 50% 34
  7. ERQA Qwen 55% 33
  8. LVBench Qwen 53% 32
  9. StrongREJECT OpenAI 67% 30
  10. RefCOCO / RefCOCO+ / RefCOCOg Qwen 50% 28
  11. INCLUDE Qwen 60% 25
  12. CountBench Qwen 58% 24
  13. AndroidWorld Qwen 61% 23
  14. DynaMath Qwen 52% 23
  15. MMLU-ProX Qwen 61% 23
  16. MLVU Qwen 77% 22
  17. MVBench Qwen 64% 22
  18. MedXpertQA-MM Qwen 52% 21
  19. LiveBench Qwen 55% 20
  20. Cybench Anthropic 58% 19
  1. BBQ (Bias Benchmark for QA) Anthropic 55% 38
  2. MRCR Google DeepMind 70% its own 37
  3. C-Eval Qwen 50% 36
  4. SuperGPQA Qwen 56% 36
  5. MMLU-Redux Qwen 71% 35
  6. RealWorldQA Qwen 57% 35
  7. Virology Capabilities Test (VCT) Anthropic 50% 34
  8. ERQA Qwen 55% 33
  9. LVBench Qwen 53% 32
  10. HealthBench OpenAI 52% its own 31
  11. Multimodal Troubleshooting Virology OpenAI 100% its own 30
  12. StrongREJECT OpenAI 67% 30
  13. Tacit Knowledge and Troubleshooting OpenAI 100% its own 29
  14. OpenAI MRCR OpenAI 50% its own 28
  15. ProtocolQA Open-Ended OpenAI 100% its own 28
  16. RefCOCO / RefCOCO+ / RefCOCOg Qwen 50% 28
  17. OpenAI Production Benchmarks and safe-completion evals OpenAI 100% its own 26
  18. INCLUDE Qwen 60% 25
  19. CountBench Qwen 58% 24
  20. Image input safety evaluations OpenAI 100% its own 24

Source: Labs’ launch posts, read by Benchmark Almanac

How long do benchmarks stay in launch posts?

Of the benchmarks labs first named in 2025, 33% appeared in only one release, and 37% were still being named a year later. Of those first named in 2026 before May 21st, 43% have appeared in only one release. This counts every lab, whatever the frontier switches say.

0% 25% 50% 75% 100% 0 3 6 9 12 15 18 21 24 Months since a lab first named it 2026 2025 2024 0% 25% 50% 75% 100% 0 3 6 9 12 15 18 21 24 Months since a lab first named it 2026 2025 2024

Source: Labs’ launch posts, read by Benchmark Almanac

How long do benchmarks stay in launch posts?

Of the benchmarks labs first named in 2025, 31% appeared in only one release, and 34% were still being named a year later. Of those first named in 2026 before May 21st, 37% have appeared in only one release. This counts every lab, whatever the frontier switches say.

0% 25% 50% 75% 100% 0 3 6 9 12 15 18 21 24 Months since a lab first named it 2026 2025 2024 0% 25% 50% 75% 100% 0 3 6 9 12 15 18 21 24 Months since a lab first named it 2026 2025 2024

Source: Labs’ launch posts, read by Benchmark Almanac

Professional work rose from 1% of what labs reported in 2025 to 10% in 2026

What each of the 16 labs with the most releases named in 2026, by area. Each grid has 100 shapes, one for every 1% of the benchmarks the lab named across its releases. Labs with similar mixes sit side by side. "Their own" is the share the lab made itself, alone or with others, and "Shared" is the share that at least three other labs also named that year.

  • Knowledge and language
  • Math and science
  • Coding
  • Agents and tools
  • Professional work
  • Vision, audio and robotics
  • Safety and security
  • Other
Google DeepMind32
Their own
21%
Shared
72%
Per release
8.5
NVIDIA11
Their own
7%
Shared
70%
Per release
28
DeepSeek8
Their own
6%
Shared
82%
Per release
10
Z.ai10
Their own
17%
Shared
72%
Per release
13.5
Microsoft6
Their own
7%
Shared
77%
Per release
8
Liquid AI7
Their own
2%
Shared
76%
Per release
11
Qwen30
Their own
5%
Shared
66%
Per release
42
Moonshot AI5
Their own
16%
Shared
76%
Per release
33
Thinking Machines5
Their own
2%
Shared
66%
Per release
27
Anthropic13
Their own
22%
Shared
58%
Per release
37
Mistral AI6
Their own
6%
Shared
69%
Per release
6
Upstage6
Their own
11%
Shared
67%
Per release
12.5
Meta9
Their own
15%
Shared
74%
Per release
17
OpenAI23
Their own
40%
Shared
60%
Per release
10
xAI7
Their own
10%
Shared
45%
Per release
6
Xiaomi Corp6
Their own
0%
Shared
77%
Per release
10.5

Source: Labs’ launch posts, read by Benchmark Almanac

Terminal-Bench was named by 33 of the 50 labs that published scores in 2026

Look up any benchmark and lab. Each dot counts the lab's releases whose launch post or documents named the benchmark, in the benchmark's area color. The columns are the 12 labs with the most releases that year, and the last one adds up every other lab. Point at a dot to see the releases and the scores they reported.

Showing the 207 benchmarks that at least 3 labs named in 2026, out of 901 in total.

Releases per lab that named each benchmark in the chosen year
BenchmarkGoogleQwenOpenAIAnthropicNVIDIAZ.aiMetaDeepSeekLiquidxAIMicrosoftMistral38 othersLabsReleases
1Terminal-BenchStanford + Laude Institute10151013659713225133/50134
2SWE-bench VerifiedPrinceton NLP + OpenAI111852222213731/5073
3tau-benchSierra Research7125283221112631/5070
4AIMEMathematical Association of America (exam), MathArena (evaluation)651244114223431/5066
5GPQA DiamondNYU (Rein et al.)82012584334214229/50112
6SWE-Bench ProScale AI4119111232123128/5077
7Humanity's Last ExamCenter for AI Safety + Scale AI10155136645213727/50104
8LiveCodeBenchUC Berkeley / MIT / Cornell7177232112526/5065
9SWE-bench MultilingualSWE-bench team (Princeton/Stanford)9124121112725/5058
10IFBenchAllenAI4179117222524/5068
11DeepSWEDatacurve46751365311823/5059
12MMLU-ProTIGER-Lab (Waterloo)616732112722/5063
13MMMU-ProMMMU team101592121112220/5064
14BrowseCompOpenAI148853222820/5061
15AutomationBenchZapier53981245211420/5054
16MCP AtlasScale AI2446134211720/5044
17SciCodeSciCode team537222220/5041
18GDPval-AAArtificial Analysis61312336222019/5058
19HMMTHarvard-MIT Mathematics Tournament (HMMT)1423212019/5042
20CharXivPrinceton NLP617172132018/5057
21OmniDocBenchShanghai AI Lab (OpenDataLab)613411111318/5040
22IFEvalGoogle512261316/5038
23ToolathlonHKUST NLP16775251415/5047
24AA-LCRArtificial Analysis1171112015/5041
25OCRBenchYuliang Liu, Xiang Bai et al.1221111114/5028
26τ³-bench BankingSierra1141211214/5022
27SimpleQAGoogle DeepMind241131914/5021
28OSWorldXLANG Lab (HKU) + collaborators611131122711113/5064
29Agents' Last ExamUC Berkeley RDI + 300+ experts5492151013/5036
30Berkeley Function Calling Leaderboard (BFCL)UC Berkeley (Gorilla)123711013/5033
31IMO-AnswerBenchGoogle DeepMind82331513/5031
32LongBenchTsinghua KEG / Zhipu113121913/5027
33SWE AtlasScale AI1211013/5014
34MATH-Vision (MathVision)CUHK MMLab613211312/5035
35ProgramBenchMeta Superintelligence Labs, Stanford University and Harvard University2822211012/5027
36DeepSearchQAGoogle DeepMind5421312/5024
37MathVistaUCLA + Microsoft11111912/5023
38MMMUMMMU team (OSU, Waterloo and others)12111811/5023
39Video-MMEVideo-MME team (USTC, XMU, et al.)16211110/5030
40BabyVisionUniPat AI + xbench1011221110/5027
41Artificial Analysis Intelligence IndexArtificial Analysis46231810/5024
42MultiChallengeScale AI9411810/5023
43MMStarUSTC and Shanghai AI Laboratory1211710/5021
44Vals Finance AgentVals AI335131510/5021
45HallusionBenchUniversity of Maryland911710/5018
46CritPtCritPt collaboration2231810/5016
47BLINKXingyu Fu et al.11910/5011
48SimpleVQAMultimodal Art Projection (M-A-P) et al.1412189/5026
49FrontierCodeCognition25811169/5024
50GDP.pdfSurge AI2341359/5018
51PostTrainBenchAISA group (ELLIS Institute Tübingen and MPI-IS) + University of Tübingen + Thoughtful Lab162189/5018
52MMLUUC Berkeley (Hendrycks et al.)141169/5013
53HealthBench ProfessionalOpenAI1112931118/5029
54ARC-AGIARC Prize Foundation376288/5026
55GraphWalksOpenAI7761138/5025
56HMMT Nov 2025MathArena (ETH Zurich SRI Lab), problems by HMMT13388/5024
57ScreenSpot-ProScreenSpot-Pro authors111322138/5023
58AI2DAllen Institute for AI (AI2)111158/5018
59JobBenchUniversity of Washington + collaborators5588/5018
60RefCOCO / RefCOCO+ / RefCOCOgUNC Chapel Hill (Licheng Yu et al.)1111148/5018
61FrontierSWEProximal21721228/5017
62OpenAI MRCROpenAI831148/5017
63Harvey LABHarvey21712128/5016
64Claw-EvalClaw-Eval team (academic)51268/5014
65PinchBenchKilo AI311278/5014
66Claw-Eval Pass^3Claw-Eval team3198/5013
67APEX-AgentsMercor11108/5012
68NL2Repo-BenchByteDance + M-A-P + 2077AI1145107/5030
69MMMLUOpenAI8114147/5028
70SuperGPQAByteDance Seed + M-A-P162267/5026
71ZeroBenchUniversity of Cambridge and others112377/5023
72MRCRGoogle DeepMind1232237/5022
73Video-MMMUNTU S-Lab (LMMs-Lab)11367/5020
74ChartographySurge AI4161337/5018
75AA-BriefcaseArtificial Analysis171267/5017
76OfficeQA ProDatabricks110147/5016
77Arena-HardArena (formerly LMArena)12197/5013
78SWE-MarathonAbundant AI22357/5012
79ERQAGoogle DeepMind1171146/5024
80WMT24++Google3115146/5024
81MedXpertQA-MMTsinghua University (TsinghuaC3I)6111126/5021
82Terminal-Bench-ScienceStanford-led community (Terminal-Bench team + domain experts)45546/5018
83MMVUYale NLP10166/5017
84WideSearchByteDance Seed7106/5017
85BioMysteryBenchAnthropic211036/5016
86CountBenchGoogle Research11146/5016
87SkillsBenchBenchFlow + 105 domain experts7166/5014
88BrowseComp-ZHResearchers from Peking University, Mindverse AI, and other groups4146/509
89GSM8KOpenAI1312116/509
90HumanEvalOpenAI13236/509
91IMOInternational Mathematical Olympiad, hosted by the Australian Maths Trust11246/508
92ScreenSpotOS-Atlas team (Shanghai AI Lab and others)111226/507
93RealWorldQAxAI17145/5022
94Codeforces ratingCodeforces741315/5016
95HealthBench HardOpenAI111125/5015
96VITA-BenchMeituan LongCat team1145/5015
97OfficeQADatabricks110125/5014
98V*Penghao Wu and Saining Xie81135/5013
99DRACOPerplexity845/5012
100FLEURSGoogle54125/5012
101MathArena (Apex, IMO, USAMO)ETH Zurich SRI Lab2465/5012
102Terminal-Bench HardArtificial Analysis4145/509
103WorldVQAMoonshot AI2165/509
104FrontierScienceOpenAI175/508
105InfographicVQAComputer Vision Center (UAB) and IIIT Hyderabad41125/508
106MATHUC Berkeley (Hendrycks et al.)31225/508
107WebArena-VerifiedServiceNow211125/507
108AdvancedIFMeta Superintelligence Labs245/506
109MMSI-BenchShanghai AI Lab (OpenRobotLab) and collaborators155/506
110WebDev ArenaArena (formerly LMArena)21125/506
111ChartQAYork University and Nanyang Technological University11125/505
112DocVQACVC, Universitat Autònoma de Barcelona and IIIT Hyderabad11125/505
113LogicVistaUCLA1135/505
114MathVerseCUHK MMLab and Shanghai AI Laboratory145/505
115PaperBenchOpenAI145/505
116LVBenchZhipu AI / Tsinghua41444/5022
117C-EvalSJTU + Tsinghua + Edinburgh141214/5018
118MMLU-ProXUniversity of Tokyo (Li Lab) and collaborators115114/5018
119DynaMathUniversity of Illinois Urbana-Champaign and UC Berkeley1144/5015
120EmbSpatialBenchFudan University1134/5014
121MMBenchShanghai AI Laboratory (OpenCompass)11124/5014
122CursorBenchCursor17234/5013
123RULERNVIDIA46114/5012
124OJBenchBeijing University of Posts and Telecommunications, Tsinghua University, and others644/5010
125BBEH micro averageGoogle DeepMind6124/509
126GDPvalOpenAI51124/509
127MATH-500OpenAI3334/509
128VoiceBenchCUHK-Shenzhen3244/509
129Artificial Analysis Coding Agent IndexArtificial Analysis14114/507
130Beyond AIMEByteDance Seed164/507
131HellaSwagUniversity of Washington + AllenAI31214/507
132MGSMGoogle Research13214/507
133MMAUUniversity of Maryland3134/507
134USAMOMathematical Association of America (contest), MathArena (evaluation)4124/507
135We-MathBeijing University of Posts and Telecommunications434/507
136Audio MCScale AI11134/506
137BigCodeBenchBigCode project12214/506
138CL-benchTencent Hunyuan (with Fudan University authors)154/506
139LMArena (Elo)Arena (formerly LMArena)31114/506
140LongVideoBenchNTU S-Lab et al.64/506
141MBPPGoogle Research334/506
142PerceptionBench (Moonshot)Moonshot AI1144/506
143ResearchRubricsScale AI64/506
144Big Bench AudioArtificial Analysis1224/505
145BIG-Bench HardSuzgun et al. (Stanford University and Google Research)1224/505
146ChartQA Provis-nlp (Masry et al.)144/505
147Chinese SimpleQAAlibaba (Taobao & Tmall)1224/505
148CVBenchNew York University (Cambrian-1 team)2124/505
149FinSearchCompByteDance Seed54/505
150LibriSpeechPanayotov et al. (Johns Hopkins University)234/505
151LiveCodeBench ProLiveCodeBench Pro team11214/505
152WildClawBenchInternLM (Shanghai AI Lab)1134/505
153Online-Mind2WebOhio State University (OSU NLP Group)11114/504
154PutnamBenchUT Austin (Tsoukalas et al.)11114/504
155WebVoyagerWebVoyager team (He et al., ACL 2024)11114/504
156MMLU-ReduxUniversity of Edinburgh16113/5018
157INCLUDEEPFL and Cohere For AI11413/5016
158MLVUBAAI + partners1423/5016
159BenchCADBenchCAD team6713/5014
160MMLongBench-DocYubo Ma et al.10213/5013
161SWE-bench MultimodalSWE-bench team (Princeton/Stanford)11023/5013
162MLE-benchOpenAI2623/5010
163VLMs Are BlindAuburn University and University of Alberta823/5010
164MCPMarkeval-sys (MCPMark team)543/509
165Multi-IFMeta723/509
166Vending-BenchAndon Labs2523/509
167Claw-Eval-MMClaw-Eval team623/508
168LAB-BenchEdison Scientific (FutureHouse)3133/507
169Vision2WebZ.ai (zai-org)4213/507
170AGIEvalMicrosoft3213/506
171Seal-0Virginia Tech423/506
172Vals IndexVals AI333/506
173Daily-OmniZiwei Zhou, Rui Wang et al.3113/505
174FrontierScience-ResearchOpenAI2123/505
175Harvey LAB-AA v1.1Artificial Analysis2123/505
176MLS-BenchMLS-Bench authors (Bohan Lyu et al.)233/505
177MotionBenchZhipu AI and Tsinghua University (THUDM)53/505
178WinoGrandeAllenAI3113/505
179Bird-SQLHKU and Alibaba DAMO Academy1213/504
180EnterpriseOps-GymServiceNow133/504
181Full-Duplex-BenchGuan-Ting Lin, Hung-yi Lee and collaborators (National Taiwan University and others)1213/504
182Global MMLU LiteCohere Labs133/504
183IMO-ProofBenchGoogle DeepMind (Superhuman Reasoning team)223/504
184Seed-TTS-evalByteDance (Seed-TTS team)223/504
185SpreadsheetBenchRenmin University (RUCKBReasoning)223/504
186SWE-Bench Pro V2 HardScale AI (co-developed with Reflection)43/504
187VisuLogicShanghai AI Laboratory with several universities43/504
188WebArenaCarnegie Mellon University2113/504
189WenetSpeechNorthwestern Polytechnical University (ASLP@NPU), Mobvoi and others223/504
190$OneMillion-BenchHumanlaya + BIGAI + xbench123/503
191AMO-BenchMeituan LongCat Team123/503
192ArtifactsBenchTencent Hunyuan123/503
193BrowseComp-VLAlibaba Tongyi Lab1113/503
194Legal Research BenchVals AI123/503
195MindCubeMLL Lab, Northwestern University, with Stanford, NYU, and University of Washington123/503
196MM-BrowseCompByteDance Seed and collaborators (Li et al.)123/503
197MM-IFEvalMM-IFEngine team (Ding et al.)123/503
198MuirBenchFei Wang, Xingyu Fu et al.123/503
199olmOCR-BenchAllen Institute for AI33/503
200PhyXHui Shen et al.123/503
201PresentBenchPresentBench team123/503
202SpreadsheetBench 2Renmin University of China (RUCKBReasoning)33/503
203TextVQAFacebook AI Research1113/503
204TriviaQAUniversity of Washington1113/503
205ViewSpatial-BenchDingming Li, Yueting Zhuang et al.123/503
206WorkSpaceBenchSJTU + ByteDance (OpenDataBox)123/503
207WritingBenchAlibaba Group (X-PLUG)33/503

How this is counted

A release is one launch post. When a lab put out several models under one post, that's one release. The system cards, model cards and technical reports that came with a post count toward the same release. A benchmark counts once per release, however many settings, subsets or documents report it, and with or without a score.

Versions of one test count as one benchmark in the graphics, so Terminal-Bench 2.0 and 3.0 are both Terminal-Bench. Tests with their own task sets stay apart, like SWE-bench Verified, SWE-Bench Pro and SWE-bench Multilingual. The lookup can show every version on its own.

A benchmark's color and shape come from its area, the same eight areas the rest of the almanac uses. Safety, alignment, bio and cyber evaluations are left out until you switch them on. Most come from system cards, and only a few labs publish those.

The quarterly chart counts 2024 as one column, because we have only 31 launch posts with scores from that year. The spread chart only shows benchmarks released since 2025, so the first lab we saw name one is the first lab that did.

The charts over time count by half year from 2024. We have read far fewer launch posts from 2024 than from later years (16 from the first half of 2024, 142 from the busiest half year since), so the earliest points rest on fewer releases. Hover a point to see how many. The latest half year isn't over yet, so its point is hollow.

A benchmark stays in launch posts until no release has named any version of it for 6 months. We can only tell that for benchmarks last named before May 21st, 2026, so a benchmark named since then counts as followed up to that day, and one first named since then doesn't count yet. Each line is a survival estimate, as for saturation on the evals page, and stops when fewer than 15 benchmarks are left to follow. The first year we read is 2024, so benchmarks labs had named before then and named again in 2024 count as first named in 2024.

"Frontier models only" counts just the releases of a model that ranked in the top ten on Epoch AI's Capabilities Index among the models out the day it was released, from any lab. It uses Epoch AI's index, so a model Epoch hasn't scored doesn't count. Both switches can be on at once.

"Frontier labs only" counts just the releases of the nine labs whose best 2026 models score highest on Epoch AI's Capabilities Index, as of November 2026. Every count on the page then compares those labs with each other, and the spread chart asks how fast a benchmark reached them.

We only count what labs reported themselves. Leaderboard and Epoch AI scores don't count here. Every number links back to its source on the benchmark's page and the model's page, and the Model cards page lists every post we read.

More insights