Benchmarks/MATH level 5

MATH level 5

MATH Level 5 — the hardest tier of the MATH benchmark, featuring competition-level problems from AMC, AIME, and Olympiad-style mathematics.

math

Models Tested

98.1

Top Score

56.9

Average Score

Rankings

#	Model	Score
1	GPT-5 ChatOpenAI	98.1
2	GPT-5OpenAI	98.1
3	GPT-5 MiniOpenAI	97.8
4	o4 MiniOpenAI	97.8
5	o3OpenAI	97.8
6	Claude Sonnet 4.5Anthropic	97.7
7	Qwen3 MaxAlibaba Qwen	97.1
8	DeepSeek-R1 (May 2025)DeepSeek	96.6
9	o3 MiniOpenAI	96.5
10	Claude Haiku 4.5Anthropic	96.4
11	GPT-5 NanoOpenAI	95.2
12	o1OpenAI	94.7
13	DeepSeek-R1DeepSeek	93.0
14	Claude 3.7 SonnetAnthropic	91.2
15	Claude 3.7 Sonnet (thinking)Anthropic	91.2
16	Grok 3 MinixAI	90.9
17	Grok 3 Mini BetaxAI	90.9
18	o1-miniOpenAI	89.2
19	Grok 3xAI	88.8
20	Grok 3 BetaxAI	88.8
21	GPT-4.1 MiniOpenAI	87.3
22	Claude Opus 4Anthropic	85.0
23	Claude Sonnet 4Anthropic	84.4
24	Gemini 2.0 ProGoogle	83.5
25	GPT-4.1OpenAI	83.0
26	o1-previewOpenAI	81.7
27	Mistral Medium 3Mistral AI	81.6
28	GPT-4.5OpenAI	78.6
29	Gemma 3 4B (free)Google DeepMind	74.0
30	Gemma 3 4BGoogle DeepMind	74.0
31	Gemma 3 12B (free)Google DeepMind	74.0
32	Gemma 3 12BGoogle DeepMind	74.0
33	Gemma 3 27B (free)Google DeepMind	74.0
34	Gemma 3 27BGoogle DeepMind	74.0
35	Llama 4 MaverickMeta	73.0
36	Gemini 1.5 Pro (Sept 2024)Google	70.4
37	GPT-4.1 NanoOpenAI	70.0
38	Qwen3 235B A22BAlibaba Qwen	68.9
39	Qwen2.5-MaxAlibaba	67.2
40	Phi 4Microsoft	64.9
41	DeepSeek V3DeepSeek	64.8
42	Grok-2 (Dec 2024)xAI	63.5
43	Qwen2.5 72B InstructAlibaba Qwen	63.2
44	Llama 4 ScoutMeta	62.3
45	Gemini 1.5 Flash (Sep 2024)Google	61.9
46	GPT-4o (2024-08-06)OpenAI	53.3
47	GPT-4o (2024-05-13)OpenAI	53.3
48	GPT-4oOpenAI	53.3
49	GPT-4o (extended)OpenAI	53.3
50	GPT-4o-mini (2024-07-18)OpenAI	52.6
51	GPT-4o-miniOpenAI	52.6
52	Claude 3.5 SonnetAnthropic	51.7
53	Mistral Large 2411Mistral AI	50.3
54	GPT-4o (2024-11-20)OpenAI	49.8
55	Llama 3.1-405BMeta	49.8
56	GPT-4 Turbo PreviewOpenAI	46.7
57	Claude 3.5 HaikuAnthropic	46.4
58	Mistral Large 2407Mistral AI	44.8
59	Llama 3.3 70B Instruct (free)Meta	41.6
60	Llama 3.3 70B InstructMeta	41.6
61	Gemini 1.5 Pro (Feb 2024)Google	40.8
62	Llama 3.2 90BMeta	39.4
63	Qwen2-72BAlibaba	39.1
64	Claude 3 OpusAnthropic	37.5
65	Llama 3.1 70B InstructMeta	36.7
66	Gemma 2 27BGoogle DeepMind	27.9
67	Gemini 1.5 Flash (May 2024)Google	25.1
68	Mistral LargeMistral AI	24.5
69	Mixtral 8x22B InstructMistral AI	24.2
70	GPT-4 TurboOpenAI	23.0
71	GPT-4 Turbo (older v1106)OpenAI	23.0
72	GPT-4 (older v0314)OpenAI	23.0
73	GPT-4OpenAI	23.0
74	Llama 3.1 8B InstructMeta	22.9
75	Llama 3 70B InstructMeta	22.6
76	Gemma 2 9BGoogle DeepMind	21.0
77	Claude 3 SonnetAnthropic	18.2
78	phi-3-medium 14BMicrosoft	17.6
79	Claude 3 HaikuAnthropic	14.9
80	Claude 2Anthropic	11.7
81	GPT-3.5 Turbo (older v0613)OpenAI	11.6
82	GPT-3.5 Turbo InstructOpenAI	11.6
83	GPT-3.5 TurboOpenAI	11.6
84	Gemini 1.0 ProGoogle	11.2
85	Mistral NemoMistral AI	10.8
86	Mixtral 8x7B InstructMistral AI	9.9
87	Llama 3 8B InstructMeta	6.1
88	U Yi-34Bunknown	5.2
89	Llama 2-70BMeta	3.3

Related Benchmarks

GSM8K

48 models

OTIS Mock AIME 2024-2025

105 models

FrontierMath-2025-02-28-Private

60 models

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

39 models