Skip to content

Benchmark

PolyMATH

Math benchmarks measure how well models solve maths problems. Scores are percentages. Higher is better.

Models measured
20
Scores recorded
25
Sources
2
IQ weight
1.52

Score distribution

58.7%
Mean
57.3%
Median
19.7%
Std dev
30.4%
10th percentile
84%
90th percentile
8.2%
Min
8.2%86.5%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.52 for chat models, measured over 20 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Qwen3.7 Max· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim86.5%
2Qwen3.7 MaxQwen3.7-Max release · recommended reasoning-effort=xhigh system prompt· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.86.5%
3Qwen3.7 Plus· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim84%
4Qwen3.7 PlusQwen3.7-Plus release · recommended reasoning-effort=xhigh system prompt· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.84%
5Qwen3.6 Plus· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim77.4%
6Qwen3.6 PlusQwen official release table · thinking mode unless benchmark requires otherwise· Reported by Alibaba Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.77.4%
7Qwen3.5 397B A17B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim73.3%
8Qwen3.5 397B A17BQwen3.5 release/model card · published setting· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.73.3%
9Qwen3.5 27B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim71.2%
10Qwen3.5 122B-A10B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim68.9%
11Qwen3.5 35B A3B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim64.4%
12Qwen 3 235B A22B Thinking 2507· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim60.1%
13Qwen3.5 9B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim57.3%
14Qwen3 Next 80B A3Bthinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim56.3%
15Qwen3 VL 32B Thinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.52%
16qwen3-vl-30bthinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.51.7%
17Qwen3.5 4B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim51.1%
18Qwen3 235B A22B Instruct 2507· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim50.2%
19Qwen: Qwen3 VL 8B Thinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.47.5%
20Qwen3 Next 80B A3B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim45.9%
21qwen3-vl-30b· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.44.3%
22Qwen3 VL 32B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.40.5%
23Qwen3 VL 8B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it.30.4%
24Qwen3.5 2B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim26.1%
25Qwen3.5 0.8B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim8.2%

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.