Multilingual

PolyMath leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

As of 2026-10-10, the highest measured score on PolyMath is 86.5% by Qwen3.7 Max. 12 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Qwen3.7 Max	86.5%	measured
2	Qwen3.7 Plus	84.0%	measured
3	Claude Opus 4.5	79.0%	measured
4	Qwen3.6 Plus	73.8%	estimated ± 11.4 pp, low confidence
5	Qwen3.5 397B	73.3%	measured
6	K-EXAONE 2.0	71.3%	measured
7	GLM-5	50.1%	estimated ± 11.4 pp, low confidence
8	Nemotron 3 Ultra	49.0%	estimated ± 11.4 pp, low confidence
9	Kimi K2.5	43.1%	measured
10	Qwen3.5-122B-A10B	42.5%	estimated ± 11.4 pp, low confidence
11	Qwen3.5-27B	42.5%	estimated ± 11.4 pp, low confidence
12	Qwen3 235B 2507 (Reasoning)	39.2%	estimated ± 11.4 pp, low confidence
13	Qwen3.5-35B-A3B	39.2%	estimated ± 11.4 pp, low confidence
14	Qwen3 235B 2507	38.4%	estimated ± 11.4 pp, low confidence
15	GPT-4.1	38.2%	estimated ± 11.4 pp, low confidence
16	DeepSeek V3 0324	38.2%	estimated ± 11.4 pp, low confidence
17	GPT-4o	38.2%	estimated ± 11.4 pp, low confidence
18	Phi-4	38.2%	estimated ± 11.4 pp, low confidence

PolyMath leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General