Benchmark
PHYBench
Science benchmarks measure how well models answer science questions. Scores are percentages. Higher is better.
- Models measured
- 29
- Scores recorded
- 32
- Sources
- 3
- IQ weight
- 1.32
Score distribution
- 39%
- Mean
- 37.9%
- Median
- 15.4%
- Std dev
- 23.7%
- 10th percentile
- 56.5%
- 90th percentile
- 15.3%
- Min
This benchmark has a weight of 1.32 for chat models, measured over 29 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 | Hy3 (Deep Infra)· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 77.4% |
| 2 | Hy3 (Deep Infra)Hy3 official Benchmark Appendix · highest reasoning tier unless noted· Reported by Tencent Hunyuan· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 77.4% |
| 3 | Gemini 3 Pro Previewthinking· PHYBench | 61.4% |
| 4 | Gemini 2.5 Pro· PHYBench | 56.5% |
| 5 | Seed 2.0 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 56% |
| 6 | Grok 4· PHYBench | 53% |
| 7 | GPT 5· PHYBench | 51.1% |
| 8 | OpenAI o3high· PHYBench | 46.4% |
| 9 | Doubao-Seed 1.6· PHYBench | 45.5% |
| 10 | Qwen 3 235B A22B Thinking 2507· PHYBench | 45.4% |
| 11 | Gemini 2.5 Flash· PHYBench | 44.8% |
| 12 | DeepSeek R1· PHYBench | 44.3% |
| 13 | o4 Mini· PHYBench | 41.9% |
| 14 | Doubao Seed 1.8· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 41% |
| 15 | Qwen 3 235B A22B· PHYBench | 40.8% |
| 16 | GLM 4.5· PHYBench | 37.9% |
| 17 | o3 Mini· PHYBench | 37.3% |
| 18 | Qwen3 Max Preview· PHYBench | 33.6% |
| 19 | kimi-k2-0711-preview· PHYBench | 33.6% |
| 20 | Grok 3beta· PHYBench | 32% |
| 21 | Claude Sonnet 4thinking· PHYBench | 30.8% |
| 22 | DeepSeek V3.1· PHYBench | 27.5% |
| 23 | o1· PHYBench | 27.4% |
| 24 | QwQ 32B· PHYBench | 27.4% |
| 25 | Claude Sonnet 3.7thinking· PHYBench | 27.1% |
| 26 | o3 Minilow· PHYBench | 25.4% |
| 27 | DeepSeek V3· PHYBench | 24.2% |
| 28 | Claude Sonnet 3.7· PHYBench | 23.8% |
| 29 | GPT 4.1· PHYBench | 23.7% |
| 30 | DeepSeek R1 Distill Qwen 32B· PHYBench | 22.1% |
| 31 | GPT 4o· PHYBench | 15.4% |
| 32 | Qwen Max· PHYBench | 15.3% |
Data credits: Data by LLM Stats · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.