Epoch AI - Critpt: leaderboard
CritPt poses unpublished physics research problems: 71 composite challenges written by active researchers, also split into about 190 checkpoint tasks, with final answers that an automated grader checks. Epoch AI does not run it and republishes the accuracy that Artificial Analysis measures on the 70 test challenges, higher is better. Source: artificialanalysis.ai/evaluations/critpt.
Metric: Score. Source: epoch.ai. Saturation forecast: Around June 2028. 232 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (Max) | 32.29 | #19 (GPT-5.6 Sol) |
| 2 | GPT-6 Astra (Max) | 31.71 | #6 (GPT-6 Astra) |
| 3 | GPT-6.1 Sol (Max) | 31.71 | #7 (GPT-6.1 Sol) |
| 4 | GPT-6.1 Sol (xHigh) | 31.71 | #7 (GPT-6.1 Sol) |
| 5 | Claude Opus 5.5 (Max) | 31.71 | #3 (Claude Opus 5.5) |
| 6 | Claude Opus 5.5 (xHigh) | 31.71 | #3 (Claude Opus 5.5) |
| 7 | GPT-6 Astra (xHigh) | 31.43 | #6 (GPT-6 Astra) |
| 8 | Claude Sonnet 5.5 (Max) | 31.43 | #11 (Claude Sonnet 5.5) |
| 9 | Claude Sonnet 5.5 (xHigh) | 31.14 | #11 (Claude Sonnet 5.5) |
| 10 | Claude Fable 5.1 (xHigh) | 31.14 | #9 (Claude Fable 5.1) |
| 11 | GPT-6 Sol (Max) | 30.86 | #14 (GPT-6 Sol) |
| 12 | Claude Opus 5.5 (High) | 30.86 | #3 (Claude Opus 5.5) |
| 13 | GPT-5.5 Pro (xHigh) | 30.57 | #17 (GPT-5.5 Pro) |
| 14 | Claude Fable 5.1 (High) | 30.29 | #9 (Claude Fable 5.1) |
| 15 | GPT-5.6 Terra (Max) | 30 | #41 (GPT-5.6 Terra) |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=epoch-ai-critpt · How It Works · Data refreshed daily, snapshot 2026-10-20.