Sources & licences
Every score links back to where it was published. These are the datasets we pull from.
Epoch AI Benchmarking Hub
CC BY 4.0
2,941 records · fetched Nov 20, 2026
models.dev
MIT
527 records · fetched Nov 20, 2026
OpenRouter models + endpoints API
Public API, attributed
1,266 records · fetched Nov 20, 2026
Epoch AI citation: Epoch AI, ‘Capabilities & benchmarking’. Published online at epoch.ai. Retrieved from https://epoch.ai/benchmarks.
Benchmark origins
| Benchmark | Results come from | Models | Official site |
|---|---|---|---|
| SWE-bench Verified | Run by Epoch AI | 32 | www.swebench.com |
| Terminal-Bench 2.0 | Benchmark leaderboard, collected by Epoch AI | 44 | www.tbench.ai |
| DeepSWE | Benchmark leaderboard, collected by Epoch AI | 26 | epoch.ai |
| FrontierSWE | Benchmark leaderboard, collected by Epoch AI | 25 | epoch.ai |
| CursorBench | Benchmark leaderboard, collected by Epoch AI | 22 | cursor.com |
| GSO-Bench | Benchmark leaderboard, collected by Epoch AI | 30 | gso-bench.github.io |
| METR Time Horizons | Benchmark leaderboard, collected by Epoch AI | 35 | metr.org |
| Vending-Bench 2 | Benchmark leaderboard, collected by Epoch AI | 68 | andonlabs.com |
| PostTrainBench | Benchmark leaderboard, collected by Epoch AI | 21 | epoch.ai |
| APEX-Agents | Benchmark leaderboard, collected by Epoch AI | 62 | www.mercor.com |
| GDPval | Benchmark leaderboard, collected by Epoch AI | 11 | openai.com |
| Remote Labor Index | Benchmark leaderboard, collected by Epoch AI | 18 | www.remotelabor.ai |
| DeepResearch Bench | Benchmark leaderboard, collected by Epoch AI | 24 | deepresearch-bench.github.io |
| OSWorld 2.0 | Benchmark leaderboard, collected by Epoch AI | 9 | os-world.github.io |
| OSWorld | Benchmark leaderboard, collected by Epoch AI | 9 | os-world.github.io |
| Cybench | Benchmark leaderboard, collected by Epoch AI | 21 | cybench.github.io |
| CL-bench | Benchmark leaderboard, collected by Epoch AI | 19 | epoch.ai |
| TheAgentCompany | Benchmark leaderboard, collected by Epoch AI | 14 | the-agent-company.com |
| ExploitBench | Benchmark leaderboard, collected by Epoch AI | 9 | epoch.ai |
| BALROG | Benchmark leaderboard, collected by Epoch AI | 37 | balrogai.com |
| Aider Polyglot | Benchmark leaderboard, collected by Epoch AI | 45 | aider.chat |
| MirrorCode | Run by Epoch AI | 41 | epoch.ai |
| FrontierCode | Benchmark leaderboard, collected by Epoch AI | 46 | epoch.ai |
| WebDev Arena | Benchmark leaderboard, collected by Epoch AI | 104 | web.lmarena.ai |
| SciCode | Benchmark leaderboard, collected by Epoch AI | 139 | scicode-bench.github.io |
| WeirdML v3 | Benchmark leaderboard, collected by Epoch AI | 27 | htihle.github.io |
| AlgoTune | Benchmark leaderboard, collected by Epoch AI | 18 | algotune.io |
| GPQA Diamond | Run by Epoch AI | 216 | arxiv.org |
| Humanity's Last Exam | Benchmark leaderboard, collected by Epoch AI | 43 | lastexam.ai |
| ARC-AGI-2 | Benchmark leaderboard, collected by Epoch AI | 96 | arcprize.org |
| ARC-AGI | Benchmark leaderboard, collected by Epoch AI | 95 | arcprize.org |
| SimpleBench | Benchmark leaderboard, collected by Epoch AI | 100 | simple-bench.com |
| Chess Puzzles | Run by Epoch AI | 94 | epoch.ai |
| CritPt | Benchmark leaderboard, collected by Epoch AI | 162 | epoch.ai |
| FrontierMath Tiers 1-3 | Run by Epoch AI | 94 | epoch.ai |
| FrontierMath Tier 4 | Run by Epoch AI | 76 | epoch.ai |
| OTIS Mock AIME | Run by Epoch AI | 202 | epoch.ai |
| MATH Level 5 | Run by Epoch AI | 83 | github.com |
| ProofBench | Benchmark leaderboard, collected by Epoch AI | 84 | epoch.ai |
| SimpleQA Verified | Run by Epoch AI | 95 | epoch.ai |
| MMLU | Benchmark leaderboard, collected by Epoch AI | 116 | arxiv.org |
| LMCA | Benchmark leaderboard, collected by Epoch AI | 140 | epoch.ai |
| DTBench | Benchmark leaderboard, collected by Epoch AI | 172 | epoch.ai |
| Fiction.LiveBench | Benchmark leaderboard, collected by Epoch AI | 51 | fiction.live |
| VPCT | Benchmark leaderboard, collected by Epoch AI | 24 | epoch.ai |
| GeoBench | Benchmark leaderboard, collected by Epoch AI | 42 | epoch.ai |