Skip to content

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
15.3%77.4%
The band covers the middle 80% of scores. The line marks the median.

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

#ModelScore
1Hy3 (Deep Infra)· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim77.4%
2Hy3 (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%
3Gemini 3 Pro Previewthinking· PHYBench61.4%
4Gemini 2.5 Pro· PHYBench56.5%
5Seed 2.0 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim56%
6Grok 4· PHYBench53%
7GPT 5· PHYBench51.1%
8OpenAI o3high· PHYBench46.4%
9Doubao-Seed 1.6· PHYBench45.5%
10Qwen 3 235B A22B Thinking 2507· PHYBench45.4%
11Gemini 2.5 Flash· PHYBench44.8%
12DeepSeek R1· PHYBench44.3%
13o4 Mini· PHYBench41.9%
14Doubao Seed 1.8· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim41%
15Qwen 3 235B A22B· PHYBench40.8%
16GLM 4.5· PHYBench37.9%
17o3 Mini· PHYBench37.3%
18Qwen3 Max Preview· PHYBench33.6%
19kimi-k2-0711-preview· PHYBench33.6%
20Grok 3beta· PHYBench32%
21Claude Sonnet 4thinking· PHYBench30.8%
22DeepSeek V3.1· PHYBench27.5%
23o1· PHYBench27.4%
24QwQ 32B· PHYBench27.4%
25Claude Sonnet 3.7thinking· PHYBench27.1%
26o3 Minilow· PHYBench25.4%
27DeepSeek V3· PHYBench24.2%
28Claude Sonnet 3.7· PHYBench23.8%
29GPT 4.1· PHYBench23.7%
30DeepSeek R1 Distill Qwen 32B· PHYBench22.1%
31GPT 4o· PHYBench15.4%
32Qwen Max· PHYBench15.3%

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.