AA CritPt v2: leaderboard
Artificial Analysis independently evaluates 70 revised research-level physics challenges with code execution in an offline Linux sandbox. CritPt authors and physics experts corrected underspecified statements and errors. Requested effort and default-fallback configurations remain separate. The earlier CritPt score is a distinct protocol.
Metric: Accuracy (%). Source: artificialanalysis.ai. Saturation forecast: Around February 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-6 Astra (Max) | 83.14 | #6 (GPT-6 Astra) |
| 2 | GPT-6.1 Sol (Max) | 82 | #7 (GPT-6.1 Sol) |
| 3 | GPT-5.6 Sol (Max) | 74.86 | #18 (GPT-5.6 Sol) |
| 4 | Muse Wonder Full (Max) | 73.14 | #10 (Muse Wonder Full) |
| 5 | Gemini 4 Argon (High) | 72.86 | #8 (Gemini 4 Argon) |
| 6 | Muse Wonder (Max) | 69.71 | #14 (Muse Wonder) |
| 7 | Grok 4.7 (xHigh) | 50.57 | #26 (Grok 4.7) |
| 8 | DeepSeek V4.1 Flash (1017) (Max) | 50 | #20 (DeepSeek V4.1 Flash (1017)) |
| 9 | GPT-5 (High) | 28.86 | #104 (GPT-5) |
No result here: #4 Claude Opus 5.5, #9 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=aa-critpt-v2 · How It Works · Data refreshed daily, snapshot 2026-10-23.