Math

AIME 2024 leaderboard

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

As of 2026-10-10, the highest measured score on AIME 2024 is 95.8% by GPT-OSS 120B. 28 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	GPT-5.2	97.4%	estimated ± 2.2 pp, low confidence
2	Step 3.5 Flash	96.3%	estimated ± 2.2 pp, low confidence
3	MAI-Thinking-1	96.1%	estimated ± 2.2 pp, low confidence
4	GPT-OSS 120B	95.8%	measured
5	Kimi K2.5	95.7%	estimated ± 2.2 pp, low confidence
6	Kimi K2.5 (Reasoning)	95.7%	estimated ± 2.2 pp, low confidence
7	GLM-4.7	95.6%	estimated ± 2.2 pp, low confidence
8	Ternary Bonsai 2 27B	95.3%	estimated ± 2.2 pp, low confidence
9	MiMo-V2-Flash	94.9%	estimated ± 2.2 pp, low confidence
10	DeepSeek V3.2 (Thinking)	94.4%	estimated ± 2.2 pp, low confidence
11	o4-mini (high)	94.2%	estimated ± 2.2 pp, low confidence
12	Qwen3 235B 2507 (Reasoning)	94.0%	estimated ± 2.2 pp, medium confidence
13	GLM-4.7-Flash	93.7%	estimated ± 2.2 pp, medium confidence
14	DeepSeek V3.1 (Reasoning)	93.1%	measured
15	Nemotron 3 Super 100B	93.1%	estimated ± 2.2 pp, medium confidence
16	Granite 4.2 30B	92.6%	estimated ± 2.2 pp, medium confidence
17	Nemotron 3 Nano 30B	92.6%	estimated ± 2.2 pp, medium confidence
18	Sarvam 105B	92.2%	estimated ± 2.2 pp, medium confidence
19	GPT-OSS 20B	92.1%	measured
20	Gemini 2.5 Pro	92.0%	measured
21	Claude Sonnet 4.5	91.5%	estimated ± 2.2 pp, medium confidence
22	Granite 4.2 8B	91.4%	estimated ± 2.2 pp, medium confidence
23	MiniCPM5-2B	91.3%	estimated ± 2.2 pp, medium confidence
24	Exaone 4.0 32B	90.7%	estimated ± 2.2 pp, medium confidence
25	Ministral 3 14B (Reasoning)	89.8%	measured
26	Grok 3 Mini	89.5%	measured
27	Nemotron 3 Nano Omni 30B A3B	89.1%	estimated ± 2.2 pp, medium confidence
28	o3-mini	87.3%	measured
29	Granite 4.2 3B	87.1%	estimated ± 2.2 pp, medium confidence
30	MiniMax M1 80k	86.0%	measured
31	Nemotron Ultra 253B	83.8%	estimated ± 2.2 pp, medium confidence
32	Qwen3 235B 2507	82.5%	estimated ± 2.2 pp, medium confidence
33	DeepSeek-R1	79.8%	measured
34	LFM2.5-2.6B	69.7%	estimated ± 2.2 pp, medium confidence
35	Kimi K2	69.6%	measured
36	DeepSeek V3.1	66.3%	measured
37	LFM2.5-8B-A1B	61.6%	estimated ± 2.2 pp, low confidence
38	MiniCPM5-1B	59.7%	estimated ± 2.2 pp, low confidence
39	Grok 3 [Beta]	52.2%	measured
40	Claude 4 Sonnet	52.1%	estimated ± 2.2 pp, low confidence
41	DeepSeek V3	39.2%	measured

AIME 2024 leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General