PHYBench: leaderboard
Five hundred original physics problems from high school to Olympiad level (mechanics, electromagnetism, thermodynamics, optics, modern physics), scored by expression edit distance to the reference solution.
Metric: EED Score. Source: www.phybench.cn. Saturation forecast: Around December 2026. 32 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 56.54 | #152 |
| 2 | Grok 4 | 53.03 | #181 |
| 3 | GPT-5 | 51.07 | #98 |
| 4 | Gemini 2.5 Pro (Preview 03-25) | 49.46 | #123 |
| 5 | O3 (High) | 46.37 | #124 (O3) |
| 6 | Doubao-Seed-1.6 (Thinking) | 45.53 | |
| 7 | Gemini 2.5 Flash | 44.84 | #245 |
| 8 | DeepSeek R1 | 44.25 | #251 |
| 9 | O4 Mini (High) | 41.94 | #180 (O4 Mini) |
| 10 | GLM-4.5 | 37.86 | #275 |
| 11 | O3 Mini (High) | 37.26 | #272 (O3 Mini) |
| 12 | O4 Mini | 36.44 | #180 |
| 13 | Qwen 3 Max | 33.64 | #208 |
| 14 | Kimi K2 | 33.56 | #243 |
| 15 | O3 Mini | 33.26 | #272 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=phybench · How It Works · Data refreshed daily, snapshot 2026-10-20.