FrontierPhysics v0.1 Pass@3
Science
Share of the 56 FrontierPhysics v0.1 physics research tasks that a configuration passes, under the headline Overall Pass rule, in at least one of its three trials. It comes from the same rollouts as the headline and is a component of its factor group.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Opus 5.5 | 37.5 |
| GPT-6 Astra | 26.8 |
| GPT-6.1 Sol | 25.0 |
| GPT-6 Sol | 23.2 |
| Claude Fable 5.1 | 19.6 |
| Claude Opus 5 | 17.9 |
| Gemini 4 Argon | 16.1 |
| Grok 4.7 | 14.3 |
| GPT-5.6 Sol | 12.5 |
| Claude Haiku 5.5 | 10.7 |
| GPT-5.6 Terra | 5.4 |
| Grok 4.6 | 5.4 |
| GLM-5.3 | 3.6 |
| GPT-5.6 Luna | 3.6 |
| MiMo-V2.6-Pro | 3.6 |
| Kimi K3 | 1.8 |
| Mistral Large 4 Preview | 1.8 |
| Gemini 3.8 Flash | 0.0 |