Benchmarks/FrontierMath-2025-02-28-Private

FrontierMath-2025-02-28-Private

FrontierMath (Feb 2025) — original research-level math problems created by mathematicians, testing capabilities at the boundary of current AI mathematical reasoning.

math

Models Tested

50.0

Top Score

14.1

Average Score

Rankings

#	Model	Score
1	GPT-5.4 ProOpenAI	50.0
2	GPT-5.4OpenAI	47.6
3	Claude Opus 4.6Anthropic	40.7
4	GPT-5.2 ChatOpenAI	40.7
5	GPT-5.2OpenAI	40.7
6	Gemini 3 ProGoogle	37.6
7	Gemini 3.1 Pro PreviewGoogle DeepMind	36.9
8	Gemini 3 Flash PreviewGoogle DeepMind	35.6
9	GPT-5 ChatOpenAI	32.4
10	GPT-5OpenAI	32.4
11	Claude Sonnet 4.6Anthropic	32.4
12	GPT-5.1OpenAI	31.0
13	GPT-5.1 ChatOpenAI	31.0
14	M Kimi K2.5moonshotai	27.9
15	GPT-5 MiniOpenAI	27.2
16	o4 MiniOpenAI	24.8
17	DeepSeek V3.2DeepSeek	22.1
18	M Kimi K2 Thinkingmoonshotai	21.4
19	Claude Opus 4.5Anthropic	20.7
20	Grok 4xAI	19.7
21	o3OpenAI	18.7
22	ZA GLM 5 Turboz-ai	16.4
23	ZA GLM 5z-ai	16.4
24	Claude Sonnet 4.5Anthropic	15.2
25	Gemini 2.5 ProGoogle DeepMind	14.1
26	o3 MiniOpenAI	12.4
27	o1OpenAI	9.3
28	Qwen3 235B A22B Thinking 2507Alibaba Qwen	8.5
29	GPT-5 NanoOpenAI	8.3
30	Claude Opus 4.1Anthropic	7.2
31	Claude Haiku 4.5Anthropic	5.9
32	Grok 3 MinixAI	5.9
33	Grok 3 Mini BetaxAI	5.9
34	GPT-4.1OpenAI	5.5
35	Claude Opus 4Anthropic	4.5
36	GPT-4.1 MiniOpenAI	4.5
37	Claude Sonnet 4Anthropic	4.1
38	Claude 3.7 SonnetAnthropic	4.1
39	Claude 3.7 Sonnet (thinking)Anthropic	4.1
40	ZA GLM 4.6z-ai	3.8
41	Grok 3xAI	3.8
42	Grok 3 BetaxAI	3.8
43	ZA GLM 4.7z-ai	2.4
44	DeepSeek V3DeepSeek	1.7
45	o1-miniOpenAI	1.7
46	GPT-4.1 NanoOpenAI	1.0
47	Claude 3.5 SonnetAnthropic	1.0
48	Qwen2.5-MaxAlibaba	1.0
49	Llama 4 MaverickMeta	0.7
50	Grok-2 (Dec 2024)xAI	0.7
51	Mistral Medium 3Mistral AI	0.3
52	GPT-4o (2024-11-20)OpenAI	0.3
53	Mistral Large 2411Mistral AI	0.3
54	Claude 3.5 HaikuAnthropic	0.3
55	GPT-4o (2024-08-06)OpenAI	0.3
56	GPT-4o (2024-05-13)OpenAI	0.3
57	GPT-4oOpenAI	0.3
58	GPT-4o (extended)OpenAI	0.3
59	Llama 4 ScoutMeta	0.1
60	Gemini 1.5 Flash (Sep 2024)Google	0.1

Related Benchmarks

OTIS Mock AIME 2024-2025

105 models

FrontierMath-Tier-4-2025-07-01-Private

39 models