CritPt (avg@5)
Science · 2025-09-30
Artificial Analysis evaluates 70 research-level physics challenges with five independent attempts per challenge (350 scored attempts). The reported pass@1 is a repeated-run mean rather than a single 70-item proportion.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 4 Carbon | 36.0 |
| Muse Wonder Full | 32.9 |
| GPT-5.6 Sol | 32.3 |
| Claude Opus 5.5 | 31.7 |
| GPT-6.1 Sol | 31.7 |
| GPT-6 Astra | 31.7 |
| Claude Sonnet 5.5 | 31.4 |
| Claude Fable 5.1 | 31.1 |
| GPT-6 Sol | 30.9 |
| GPT-5.5 Pro | 30.6 |
| GPT-5.4 Pro | 30.0 |
| GPT-5.6 Terra | 30.0 |
| Claude Opus 5 | 29.1 |
| Claude Fable 5 | 28.6 |
| Muse Wonder | 28.3 |
| GPT-6.1 Terra | 28.0 |
| Gemini 4 Argon | 27.1 |
| GPT-5.5 | 27.1 |
| MiMo-V2.6-Pro | 26.6 |
| Ling 3.1 Flash 1019 | 26.0 |
| Muse Spark 1.3 | 26.0 |
| Gemini 3 Deep Think | 25.7 |
| Kimi K3.1 Pro | 24.6 |
| DeepSeek V4.1 Flash 1017 | 24.0 |
| GPT-5.4 | 23.4 |
| Kimi K3 | 23.4 |
| Claude Haiku 5.5 | 22.6 |
| MiniMax M3.1 Flash | 22.3 |
| Opus 4.8 | 20.9 |
| GLM-5.2 | 20.9 |
| Step 5 Preview | 20.9 |
| GPT-5.6 Luna | 20.6 |
| Qwen3.8 2.4T A95B | 20.0 |
| Qwen3.8-Max | 20.0 |
| Grok 4.6 | 19.7 |
| GPT-6 Luna | 19.4 |
| GLM-5.3 | 19.1 |
| Gemini 3.8 Flash | 18.3 |
| DeepSeek V4 Pro 0813 | 18.0 |
| Grok 4.7 | 18.0 |