Benchmark
PHYBench Accuracy
Science benchmarks measure how well models answer science questions. Scores are percentages. Higher is better.
- Models measured
- 26
- Scores recorded
- 28
- Sources
- 1
- IQ weight
- 1.55
Score distribution
- 24.5%
- Mean
- 22.1%
- Median
- 11.7%
- Std dev
- 12.9%
- 10th percentile
- 40.4%
- 90th percentile
- 6.3%
- Min
This benchmark has a weight of 1.55 for chat models, measured over 26 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Pro Previewthinking· PHYBench | 50.4% |
| 2 | Gemini 2.5 Pro· PHYBench | 45.8% |
| 3 | Grok 4· PHYBench | 42.3% |
| 4 | GPT 5· PHYBench | 40.4% |
| 5 | Qwen 3 235B A22B Thinking 2507· PHYBench | 35% |
| 6 | OpenAI o3high· PHYBench | 34.8% |
| 7 | Doubao-Seed 1.6· PHYBench | 34.5% |
| 8 | DeepSeek R1· PHYBench | 33.5% |
| 9 | Gemini 2.5 Flash· PHYBench | 33.5% |
| 10 | o4 Mini· PHYBench | 29.4% |
| 11 | Qwen 3 235B A22B· PHYBench | 28.7% |
| 12 | GLM 4.5· PHYBench | 26% |
| 13 | o3 Mini· PHYBench | 25% |
| 14 | Qwen3 Max Preview· PHYBench | 22.1% |
| 15 | kimi-k2-0711-preview· PHYBench | 21.3% |
| 16 | Grok 3beta· PHYBench | 21.2% |
| 17 | Claude Sonnet 4thinking· PHYBench | 19.2% |
| 18 | o1· PHYBench | 18% |
| 19 | DeepSeek V3.1· PHYBench | 16.6% |
| 20 | QwQ 32B· PHYBench | 15.9% |
| 21 | Claude Sonnet 3.7thinking· PHYBench | 15.3% |
| 22 | o3 Minilow· PHYBench | 13.7% |
| 23 | DeepSeek V3· PHYBench | 13.6% |
| 24 | GPT 4.1· PHYBench | 13.2% |
| 25 | Claude Sonnet 3.7· PHYBench | 12.9% |
| 26 | DeepSeek R1 Distill Qwen 32B· PHYBench | 11.7% |
| 27 | GPT 4o· PHYBench | 7% |
| 28 | Qwen Max· PHYBench | 6.3% |