Vision & documents

We-Math leaderboard

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

As of 2026-10-10, the highest measured score on We-Math is 87.9% by Qwen3.5 397B. 62 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Qwen3.8 Max	97.3%	estimated ± 1.9 pp, low confidence
2	Kimi K3	96.1%	estimated ± 1.9 pp, low confidence
3	Seed 2.1 Pro	93.9%	estimated ± 1.9 pp, low confidence
4	Qwen3.8-Omni-Flash	92.8%	estimated ± 1.9 pp, low confidence
5	Qwen3.8-Flash-Next	91.2%	estimated ± 1.9 pp, low confidence
6	Qwen3.7 Plus	90.8%	estimated ± 1.9 pp, low confidence
7	Seed 2.1 Turbo	90.6%	estimated ± 1.9 pp, low confidence
8	Qwen3.8-27B	90.5%	estimated ± 1.9 pp, low confidence
9	Qwen3.5 397B	87.9%	measured
10	Qwen3.6 Plus	87.8%	estimated ± 1.9 pp, medium confidence
11	dots3-note Preview	87.4%	estimated ± 1.9 pp, medium confidence
12	Kimi K2.6	87.0%	estimated ± 1.9 pp, medium confidence
13	Gemini 3 Pro	86.9%	measured
14	Gemini 3.1 Pro	85.5%	estimated ± 6.2 pp, low confidence
15	Qwen3.5-122B-A10B	85.5%	estimated ± 1.9 pp, medium confidence
16	Gemini 3.5 Flash	85.3%	estimated ± 6.2 pp, low confidence
17	MiMo-V2.5	85.3%	estimated ± 3.7 pp, medium confidence
18	Qwen3.6-27B	85.3%	estimated ± 3.7 pp, medium confidence
19	Qwen3.5-27B	85.2%	estimated ± 1.9 pp, medium confidence
20	GPT-5.6 Sol	85.0%	estimated ± 6.2 pp, low confidence
21	Kimi K2.5	84.7%	measured
22	Claude Mythos 5	84.6%	estimated ± 7.9 pp, low confidence
23	Claude Opus 4.7 (Adaptive)	84.6%	estimated ± 7.9 pp, low confidence
24	Claude Opus 4.8	84.6%	estimated ± 7.9 pp, low confidence
25	GLM-5.3-Flash	84.6%	estimated ± 7.9 pp, low confidence
26	Gemini 3.7 Flash	84.6%	estimated ± 7.9 pp, low confidence
27	Muse Spark 1.1	84.6%	estimated ± 7.9 pp, low confidence
28	Claude Sonnet 5	84.6%	estimated ± 7.9 pp, low confidence
29	Sakana Fugu-Ultra	84.6%	estimated ± 7.9 pp, low confidence
30	Sakana Fugu	84.6%	estimated ± 7.9 pp, low confidence
31	Claude Sonnet 4.6	84.6%	estimated ± 7.9 pp, low confidence
32	Gemini 3.1 Flash-Lite	84.5%	estimated ± 7.9 pp, low confidence
33	GPT-5.4	83.9%	estimated ± 6.2 pp, low confidence
34	GPT-5.5	83.9%	estimated ± 6.2 pp, low confidence
35	Qwen3.6-35B-A3B	83.7%	estimated ± 3.7 pp, medium confidence
36	GPT-5.6 Terra	83.6%	estimated ± 6.2 pp, low confidence
37	Muse Spark	83.4%	estimated ± 6.2 pp, low confidence
38	Qwen3.5-35B-A3B	82.4%	estimated ± 1.9 pp, medium confidence
39	Kimi K2.5 (Reasoning)	82.3%	estimated ± 6.2 pp, low confidence
40	GPT-5.6 Luna	82.2%	estimated ± 6.2 pp, low confidence
41	Grok 4.3	82.0%	estimated ± 6.2 pp, low confidence
42	Pareto 26.9	82.0%	estimated ± 6.2 pp, low confidence
43	MiniMax M3	81.9%	estimated ± 3.7 pp, medium confidence
44	Claude Opus 4.6	81.5%	estimated ± 6.2 pp, low confidence
45	Gemma 4 31B	81.3%	estimated ± 6.2 pp, low confidence
46	GPT-5.4 mini	81.1%	estimated ± 6.2 pp, low confidence
47	Step 5 Preview	80.7%	estimated ± 6.2 pp, low confidence
48	Grok 4.20	80.2%	estimated ± 6.2 pp, low confidence
49	Inkling-Small	79.4%	estimated ± 6.2 pp, low confidence
50	Muse Glimmer 30B	79.4%	estimated ± 6.2 pp, low confidence
51	Gemma 4 26B A4B	79.3%	estimated ± 6.2 pp, low confidence
52	Inkling	79.1%	estimated ± 6.2 pp, low confidence
53	GPT-5.2	79.0%	measured
54	Ternary Bonsai 2 27B	78.2%	estimated ± 5.4 pp, low confidence
55	Interfaze Beta	77.5%	estimated ± 6.2 pp, low confidence
56	Gemma 4 12B	76.9%	estimated ± 1.9 pp, medium confidence
57	Nemotron 3 Nano Omni 30B A3B	75.9%	estimated ± 5.1 pp, low confidence
58	GPT-5.4 nano	74.1%	estimated ± 6.2 pp, low confidence
59	Command A+	71.9%	estimated ± 6.2 pp, low confidence
60	ZAYA1-VL-8B	70.7%	estimated ± 4.0 pp, low confidence
61	Claude Opus 4.5	70.0%	measured
62	LFM2.5-VL-3B	69.1%	estimated ± 4.0 pp, low confidence
63	Gemma 4 E4B	63.8%	estimated ± 6.2 pp, low confidence
64	Qwen2.5-VL-32B	61.2%	estimated ± 6.2 pp, low confidence
65	Gemma 4 E2B	56.5%	estimated ± 6.2 pp, low confidence
66	North Micro Vision Instruct	43.6%	estimated ± 5.4 pp, low confidence
67	LFM2.5-VL-450M	38.7%	estimated ± 4.0 pp, low confidence

We-Math leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General