SciEval - Physics: leaderboard
SciEval multimodal scientific-reasoning leaderboard — physics-discipline accuracy.
Metric: Physics (%). Source: huggingface.co. Saturation forecast: Around 2029. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 235B A22B | 49.93 | #238 |
| 2 | Gemini 3 Pro | 45.19 | #80 |
| 3 | Qwen 3 Max | 41.04 | #208 |
| 4 | Claude Sonnet 4.5 | 40.36 | #142 |
| 5 | GPT-5 | 39.91 | #98 |
| 6 | Llama 4 Maverick | 38.97 | #464 |
| 7 | Kimi K2 | 38.59 | #243 |
| 8 | O3 | 38.14 | #124 |
| 9 | Ling-flash-2.0 | 37.98 | #550 |
| 10 | Gemini 2.5 Pro | 34.96 | #152 |
| 11 | GPT-5.1 | 32.44 | #135 |
| 12 | GPT-4o | 32.09 | #343 |
| 13 | Gemini 2.5 Flash | 32.07 | #245 |
| 14 | Intern-S1 | 31.62 | #285 |
| 15 | Grok 2 (1212) | 28.21 | #496 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=scieval-physics · How It Works · Data refreshed daily, snapshot 2026-10-20.