Math

AIME 2025 leaderboard

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

As of 2026-10-07, the highest measured score on AIME 2025 is 97.0% by MAI-Thinking-1. 23 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	GLM-5.2	99.2%	estimated ± 1.5 pp, low confidence
2	Beam	98.1%	estimated ± 1.5 pp, low confidence
3	A.X K2	97.5%	estimated ± 1.5 pp, low confidence
4	Inkling	97.5%	estimated ± 1.5 pp, low confidence
5	MAI-Thinking-1	97.0%	measured
6	Kimi K2.6	96.9%	estimated ± 1.5 pp, low confidence
7	GLM-5	96.4%	estimated ± 1.5 pp, medium confidence
8	Solar Open 2	96.3%	estimated ± 1.5 pp, medium confidence
9	Inkling-Small	96.1%	estimated ± 1.5 pp, medium confidence
10	Kimi K2.5	96.1%	measured
11	Kimi K2.5 (Reasoning)	96.1%	measured
12	GLM-5.1	95.9%	estimated ± 1.5 pp, medium confidence
13	Qwen3.6 Plus	95.9%	estimated ± 1.5 pp, medium confidence
14	Solar Pro 4	95.9%	estimated ± 1.5 pp, medium confidence
15	Claude Opus 4.5	95.8%	estimated ± 1.5 pp, medium confidence
16	GLM-4.7	95.7%	measured
17	Muse Glimmer 30B	95.4%	estimated ± 1.5 pp, medium confidence
18	Ternary Bonsai 2 27B	95.0%	measured
19	Qwen3.6-27B	94.8%	estimated ± 1.5 pp, medium confidence
20	MiMo-V2-Flash	94.1%	measured
21	Qwen3.5 397B	94.1%	estimated ± 1.5 pp, medium confidence
22	Ling 3.0 Flash	94.0%	estimated ± 1.5 pp, medium confidence
23	Qwen3.6-35B-A3B	93.5%	estimated ± 1.5 pp, medium confidence
24	K-EXAONE 2.0	93.1%	estimated ± 1.5 pp, medium confidence
25	ZAYA1-8B	89.6%	estimated ± 1.5 pp, medium confidence
26	Granite 4.2 30B	89.2%	measured
27	Claude Sonnet 4.5	87.0%	measured
28	Granite 4.2 8B	86.7%	measured
29	MiniCPM5-2B	86.5%	measured
30	Exaone 4.0 32B	85.3%	measured
31	Nemotron 3 Nano Omni 30B A3B	82.1%	measured
32	Granite 4.2 3B	78.3%	measured
33	Gemma 4 12B	74.2%	estimated ± 1.5 pp, medium confidence
34	ZAYA1-74B-Preview	72.6%	estimated ± 1.5 pp, medium confidence
35	LongCat-Flash-Lite-Sparse	57.3%	estimated ± 1.5 pp, medium confidence
36	LFM2.5-2.6B	51.9%	measured
37	LFM2.5-8B-A1B	42.5%	measured
38	MiniCPM5-1B	40.4%	measured
39	LLaDA2.2-mini	39.1%	estimated ± 1.5 pp, low confidence

AIME 2025 leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General