Benchmark

AIME

American Invitational Mathematics Examination

Competition-level mathematics problems requiring multi-step reasoning.

Models Tested

Gemini 2.5 Pro

Top Model

92%

Top Score

100 %

Max Possible

#	Model	AIME Score	Provider	Composite
1	Gemini 2.5 Pro Google	92%	Google	97.5
2	o3 OpenAI	91.6%	OpenAI	97.1
3	Claude Opus 4 Anthropic	87.4%	Anthropic	97.8
4	DeepSeek R1 DeepSeek	79.8%	DeepSeek	92.3
5	o4-mini OpenAI	79.4%	OpenAI	91.5
6	Grok 3 xAI	55.3%	xAI	88.8
7	Gemini 2.5 Flash Google	55.2%	Google	86.7
8	Llama 4 Maverick Meta	45.3%	Meta	87.5
9	Claude Sonnet 4 Anthropic	40.5%	Anthropic	90.3
10	DeepSeek V3 DeepSeek	39.2%	DeepSeek	86.1
11	Qwen 3 235B Alibaba	35.6%	Alibaba	84.5
12	Gemini 2.0 Flash Google	20.1%	Google	80.5
13	Claude 3.5 Sonnet Anthropic	16%	Anthropic	88.4
14	GPT-4o OpenAI	9.3%	OpenAI	89.2
15	Llama 3.3 70B Meta	8.2%	Meta	79.8
16	Llama 3.1 405B Meta	7.8%	Meta	82.1
17	Qwen 2.5 72B Alibaba	6.8%	Alibaba	76
18	Claude 3.5 Haiku Anthropic	6.3%	Anthropic	79.2
19	Mistral Large Mistral AI	5.6%	Mistral AI	76.5
20	Grok 2 xAI	5.2%	xAI	74.8
21	Llama 3.1 70B Meta	4.1%	Meta	74.3
22	Phi-4 Microsoft	3.8%	Microsoft	64.5
23	GPT-4o Mini OpenAI	3.2%	OpenAI	75.8
24	Mixtral 8x22B Mistral AI	2.1%	Mistral AI	65.2
25	Command R+ Cohere	1.5%	Cohere	60.8

Source: MAA, 2024