FrontierPhysics v0.1 Score
Science
FrontierPhysics v0.1 partial-credit score: each rollout that passes under the Overall Pass rule scores its weighted expert-rubric score (0-100) and every other rollout scores 0. The score is averaged over a configuration's 168 rollouts (56 tasks, three trials each). 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 | 21.3 |
| GPT-6 Astra | 16.7 |
| GPT-6.1 Sol | 15.3 |
| Claude Fable 5.1 | 12.0 |
| GPT-6 Sol | 10.0 |
| Claude Opus 5 | 9.5 |
| GPT-5.6 Sol | 7.4 |
| Gemini 4 Argon | 6.5 |
| Grok 4.7 | 6.0 |
| Claude Haiku 5.5 | 3.5 |
| GPT-5.6 Luna | 2.9 |
| Grok 4.6 | 2.6 |
| GPT-5.6 Terra | 2.5 |
| GLM-5.3 | 1.1 |
| MiMo-V2.6-Pro | 1.0 |
| Mistral Large 4 Preview | 0.5 |
| Kimi K3 | 0.4 |
| Gemini 3.8 Flash | 0.0 |