Math

FrontierMath v2 (Tiers 1-3) leaderboard

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

As of 2026-10-07, the highest measured score on FrontierMath v2 (Tiers 1-3) is 89.0% by GPT-5.6 Sol. 19 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	GPT-6 Astra	93.1%	estimated ± 7.4 pp, low confidence
2	GPT-5.6 Sol	89.0%	measured
3	GPT-5.6 Terra	84.9%	measured
4	GPT-5.6 Luna	78.6%	measured
5	GPT-5.5	51.7%	measured
6	GPT-5.5 Pro	51.0%	measured
7	GPT-5.4 Pro	50.0%	measured
8	GPT-5.4	47.6%	measured
9	Claude Opus 4.8	47.2%	measured
10	Claude Opus 4.7	43.8%	measured
11	Claude Opus 4.7 (Adaptive)	43.3%	estimated ± 0.7 pp, low confidence
12	Claude Opus 4.6	40.7%	measured
13	GPT-5.2	40.7%	measured
14	Muse Spark	39.0%	measured
15	Qwen3.7 Max	39.0%	estimated ± 8.4 pp, low confidence
16	DeepSeek V4 Flash 0731	39.0%	estimated ± 8.4 pp, low confidence
17	DeepSeek V4 Pro 0813	39.0%	estimated ± 8.4 pp, low confidence
18	Solar Open 2	39.0%	estimated ± 8.4 pp, low confidence
19	Qwen3.7 Plus	39.0%	estimated ± 8.4 pp, low confidence
20	Gemini 3.5 Flash	39.0%	measured
21	Kimi K2.6	39.0%	measured
22	GLM-5.2	39.0%	estimated ± 8.4 pp, low confidence
23	Inkling-Small	38.3%	estimated ± 8.4 pp, low confidence
24	Gemini 3 Pro	37.6%	measured
25	Gemini 3.1 Pro	36.9%	measured
26	Gemini 3 Flash	35.6%	measured
27	GLM-5.1	33.4%	measured
28	Claude Sonnet 4.6	32.4%	measured
29	GPT-5.1	31.0%	measured
30	GPT-5.4 mini	28.3%	measured
31	Kimi K2.5	27.9%	measured
32	Qwen3.5 397B	26.8%	estimated ± 8.4 pp, low confidence
33	Qwen3.6 Plus	26.2%	measured
34	GPT-5.4 nano	25.9%	measured
35	Ling 3.0 Flash	24.8%	estimated ± 8.4 pp, low confidence
36	o4-mini (high)	24.8%	measured
37	MAI-Thinking-1	24.4%	estimated ± 8.4 pp, low confidence
38	Qwen3.6-27B	24.4%	estimated ± 8.4 pp, low confidence
39	Qwen3.6-35B-A3B	24.4%	estimated ± 8.4 pp, low confidence
40	K-EXAONE 2.0	24.4%	estimated ± 8.4 pp, low confidence
41	LongCat-Flash-Lite-Sparse	24.4%	estimated ± 8.4 pp, low confidence
42	MiniCPM5-1B	24.4%	estimated ± 8.4 pp, low confidence
43	MiniCPM5-2B	24.4%	estimated ± 8.4 pp, low confidence
44	ZAYA1-8B	24.4%	estimated ± 8.4 pp, low confidence
45	Qwen 3.6 Max (preview)	23.1%	measured
46	DeepSeek V3.2	22.1%	measured
47	Kimi K2	21.4%	measured
48	Qwen3.5 Plus	21.0%	measured
49	Claude Opus 4.5	20.7%	measured
50	Grok 4	19.7%	measured
51	o3	18.7%	measured
52	GLM-5	16.4%	measured
53	Gemini 2.5 Pro	14.1%	measured
54	Claude Sonnet 4.5	13.5%	measured
55	o1	9.3%	measured
56	Qwen3 235B 2507 (Reasoning)	8.5%	measured
57	Qwen3.5 Flash	6.2%	measured
58	Claude Haiku 4.5	5.9%	measured
59	GPT-4.1	5.5%	measured
60	Gemini 2.5 Flash	4.8%	measured
61	GPT-4.1 mini	4.5%	measured
62	GLM-4.6	3.8%	measured
63	Grok 3 [Beta]	3.8%	measured
64	GLM-4.7	2.4%	measured
65	Claude 3.5 Sonnet	2.1%	measured
66	DeepSeek V3	1.7%	measured
67	GPT-4.1 nano	1.0%	measured
68	Llama 4 Maverick	0.7%	measured
69	GPT-4o	0.3%	measured
70	Llama 4 Scout	0.0%	measured

FrontierMath v2 (Tiers 1-3) leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General