Math

AIME26 leaderboard

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

As of 2026-10-07, the highest measured score on AIME26 is 99.2% by GLM-5.2. 58 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Claude Opus 4.6	100.0%	estimated ± 0.6 pp, low confidence
2	Claude Opus 4.7	100.0%	estimated ± 0.6 pp, low confidence
3	Claude Opus 4.8	100.0%	estimated ± 0.6 pp, low confidence
4	Gemini 3.1 Pro	100.0%	estimated ± 0.6 pp, low confidence
5	Gemini 3 Pro	100.0%	estimated ± 0.6 pp, low confidence
6	GPT-5.2	100.0%	estimated ± 0.6 pp, low confidence
7	GPT-5.4	100.0%	estimated ± 0.6 pp, low confidence
8	GPT-5.4 Pro	100.0%	estimated ± 0.6 pp, low confidence
9	GPT-5.5	100.0%	estimated ± 0.6 pp, low confidence
10	GPT-5.5 Pro	100.0%	estimated ± 0.6 pp, low confidence
11	GPT-5.6 Luna	100.0%	estimated ± 0.6 pp, low confidence
12	GPT-5.6 Sol	100.0%	estimated ± 0.6 pp, low confidence
13	GPT-5.6 Terra	100.0%	estimated ± 0.6 pp, low confidence
14	GPT-6 Astra	100.0%	estimated ± 0.6 pp, low confidence
15	GLM-5.2	99.2%	measured
16	Beam	97.8%	measured
17	Qwen3.7 Max	97.3%	estimated ± 1.3 pp, medium confidence
18	A.X K2	97.1%	measured
19	Inkling	97.1%	measured
20	DeepSeek V4 Pro 0813	96.9%	estimated ± 1.3 pp, medium confidence
21	DeepSeek V4 Flash 0731	96.8%	estimated ± 1.3 pp, medium confidence
22	Muse Spark	96.4%	estimated ± 0.6 pp, low confidence
23	Qwen3.7 Plus	96.4%	estimated ± 1.3 pp, high confidence
24	Gemini 3.5 Flash	96.4%	estimated ± 0.6 pp, low confidence
25	Kimi K2.6	96.4%	measured
26	Ternary Bonsai 2 27B	95.8%	measured
27	GLM-5	95.8%	measured
28	Kimi K2.5	95.8%	measured
29	Solar Open 2	95.7%	measured
30	Inkling-Small	95.5%	measured
31	GPT-5.1	95.5%	estimated ± 0.6 pp, medium confidence
32	Claude 3.5 Sonnet	95.5%	estimated ± 0.6 pp, low confidence
33	Claude Haiku 4.5	95.5%	estimated ± 0.6 pp, low confidence
34	Claude Sonnet 4.5	95.5%	estimated ± 0.6 pp, medium confidence
35	Claude Sonnet 4.6	95.5%	estimated ± 0.6 pp, medium confidence
36	DeepSeek V3.2	95.5%	estimated ± 0.6 pp, medium confidence
37	Gemini 2.5 Flash	95.5%	estimated ± 0.6 pp, medium confidence
38	Gemini 2.5 Pro	95.5%	estimated ± 0.6 pp, medium confidence
39	Gemini 3 Flash	95.5%	estimated ± 0.6 pp, medium confidence
40	GLM-4.6	95.5%	estimated ± 0.6 pp, medium confidence
41	GLM-4.7	95.5%	estimated ± 0.6 pp, low confidence
42	GPT-4.1	95.5%	estimated ± 0.6 pp, low confidence
43	GPT-5.4 mini	95.5%	estimated ± 0.6 pp, low confidence
44	GPT-5.4 nano	95.5%	estimated ± 0.6 pp, medium confidence
45	Grok 3 [Beta]	95.5%	estimated ± 0.6 pp, low confidence
46	Grok 4	95.5%	estimated ± 0.6 pp, low confidence
47	Kimi K2	95.5%	estimated ± 0.6 pp, low confidence
48	o3	95.5%	estimated ± 0.6 pp, low confidence
49	o4-mini (high)	95.5%	estimated ± 0.6 pp, medium confidence
50	Qwen3 235B 2507 (Reasoning)	95.5%	estimated ± 0.6 pp, low confidence
51	Qwen3.5 Flash	95.5%	estimated ± 0.6 pp, low confidence
52	Qwen3.5 Plus	95.5%	estimated ± 0.6 pp, low confidence
53	Qwen 3.6 Max (preview)	95.5%	estimated ± 0.6 pp, medium confidence
54	Kimi K2.5 (Reasoning)	95.4%	estimated ± 4.6 pp, medium confidence
55	o1	95.4%	estimated ± 0.6 pp, low confidence
56	DeepSeek V3	95.4%	estimated ± 0.6 pp, low confidence
57	GPT-4.1 mini	95.4%	estimated ± 0.6 pp, low confidence
58	GPT-4.1 nano	95.4%	estimated ± 0.6 pp, low confidence
59	GPT-4o	95.4%	estimated ± 0.6 pp, low confidence
60	Llama 4 Maverick	95.4%	estimated ± 0.6 pp, low confidence
61	Llama 4 Scout	95.4%	estimated ± 0.6 pp, low confidence
62	GLM-5.1	95.3%	measured
63	Qwen3.6 Plus	95.3%	measured
64	Solar Pro 4	95.3%	measured
65	Claude Opus 4.5	95.1%	measured
66	Muse Glimmer 30B	94.7%	measured
67	MAI-Thinking-1	94.5%	measured
68	Qwen3.6-27B	94.1%	measured
69	MiMo-V2-Flash	93.6%	estimated ± 4.6 pp, medium confidence
70	Qwen3.5 397B	93.3%	measured
71	Ling 3.0 Flash	93.2%	measured
72	Qwen3.6-35B-A3B	92.7%	measured
73	Granite 4.2 30B	92.4%	estimated ± 1.0 pp, low confidence
74	K-EXAONE 2.0	92.3%	measured
75	ZAYA1-8B	89.1%	measured
76	Granite 4.2 8B	87.5%	estimated ± 1.0 pp, low confidence
77	MiniCPM5-2B	86.5%	measured
78	Exaone 4.0 32B	85.5%	estimated ± 4.6 pp, medium confidence
79	Nemotron 3 Nano Omni 30B A3B	82.6%	estimated ± 4.6 pp, medium confidence
80	Granite 4.2 3B	81.4%	estimated ± 1.0 pp, low confidence
81	Gemma 4 12B	77.5%	measured
82	ZAYA1-74B-Preview	76.4%	measured
83	LongCat-Flash-Lite-Sparse	65.7%	measured
84	LFM2.5-2.6B	54.7%	estimated ± 4.6 pp, medium confidence
85	LFM2.5-8B-A1B	50.0%	measured
86	MiniCPM5-1B	40.4%	measured
87	LLaDA2.2-mini	35.1%	measured

AIME26 leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General