CritPt — leaderboard
Research-level physics reasoning benchmark with composite challenges designed by active physics researchers.
Metric: Accuracy (self-reported). Source: benchmarklist.com. Status: saturation imminent. 328 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 Pro | 30.6 |
| 2 | GPT-5.4 Pro (xHigh) | 30 |
| 3 | Claude Mythos 5 | 28.6 |
| 4 | Claude Fable 5 (Max) | 28.6 |
| 5 | GPT-5.5 | 27.1 |
| 6 | Gemini 3 Deep Think | 25.7 |
| 7 | GPT-5.4 | 23.4 |
| 8 | Claude Opus 4.8 (Max) | 20.9 |
| 9 | Gemini 3.1 Pro (Preview) | 17.7 |
| 10 | GPT-5.3 Codex | 16.9 |
| 11 | GLM-5.2 | 16.7 |
| 12 | Qwen 3.7 Max (Max) | 13.4 |
| 13 | Gemini 3.5 Flash | 13.1 |
| 14 | DeepSeek V4 Pro (Max) | 12.9 |
| 15 | Claude Opus 4.6 (Max) | 12.6 |
Interactive version: theaggregate.ai/benchmark?slug=critpt · How the rankings work · Data refreshed daily, snapshot 2026-07-22.