最近更新: 4h ago

Math 基准测试

MATH Benchmark (500-problem subset) 排行榜

Competition-level mathematics across algebra, geometry, number theory, counting/probability, intermediate algebra, and precalculus. 500-problem subset used by OpenAI for consistent evaluation.

为什么重要： Tests genuine mathematical reasoning, not just pattern matching. Reasoning models (o1, R1) dramatically outperform standard models here.

顶级模型

99%

平均评分

83.6%

共48个模型

已测试模型

指标: accuracy

人类基准

评分范围: 0%–100%

MATH-500 Scores - Top 25 Models

Ranked by MATH-500 score (%)

LMMarketCap.com

模型排名

All models with a reported MATH-500 score, ranked by highest accuracy.

排名模型评分性能

o3 OpenAI

99%

o3-mini OpenAI

97.9%

98%

97.9%

DeepSeek R1-0528 DeepSeek

97.8%

98%

97.8%

DeepSeek R1 DeepSeek

97.3%

97%

97.3%

o4-mini OpenAI

97.3%

97%

97.3%

o1 OpenAI

96.4%

96%

96.4%

Gemini 3 Pro Google

96%

GPT-5.4 OpenAI

95.5%

96%

95.5%

Gemini 2.5 Pro Google

95.2%

95%

95.2%

#10

Grok 4 xAI

95%

#11

o1 Preview OpenAI

94.8%

95%

94.8%

#12

GPT-5.2 OpenAI

94%

#13

GPT-5.1 OpenAI

93.5%

94%

93.5%

#14

GPT-5 OpenAI

92.5%

93%

92.5%

#15

DeepSeek V3 (March 2025)DeepSeek

92%

#16

Claude Opus 4.6 Anthropic

90.5%

91%

90.5%

#17

DeepSeek V3 DeepSeek

90.2%

90%

90.2%

#18

o1-mini OpenAI

90%

#19

Gemini 2.0 Flash Google

89.7%

90%

89.7%

#20

Claude Opus 4.5 Anthropic

88.1%

88%

88.1%

#21

Gemini 3 Flash Google

88%

#22

Claude Opus 4 Anthropic

86%

#23

Gemini 2.5 Flash Google

85.8%

86%

85.8%

#24

Claude Sonnet 4.6 Anthropic

85.3%

85%

85.3%

#25

Grok 3 xAI

85%

#26

Qwen 2.5 Coder 32B Alibaba

83.5%

84%

83.5%

#27

Qwen 2.5 72B Alibaba

83.1%

83%

83.1%

#28

Claude Sonnet 4.5 Anthropic

83%

#29

Claude 3.7 Sonnet Anthropic

82.2%

82%

82.2%

#30

Claude Sonnet 4 Anthropic

81.4%

81%

81.4%

#31

Llama 4 Maverick Meta

81%

#32

Phi-4 Microsoft

80.4%

80%

80.4%

#33

GPT-4.1 OpenAI

78.5%

79%

78.5%

#34

Claude 3.5 Sonnet Anthropic

78.3%

78%

78.3%

#35

Llama 3.3 70B Meta

77%

#36

GPT-4o OpenAI

76.6%

77%

76.6%

#37

Grok 2 xAI

76.1%

76%

76.1%

#38

Mistral Large 2 Mistral AI

76%

#39

Llama 3.1 405B Meta

73.8%

74%

73.8%

#40

GPT-4 Turbo OpenAI

72.6%

73%

72.6%

#41

Claude Haiku 4.5 Anthropic

72.5%

73%

72.5%

#42

GPT-4o mini OpenAI

70.2%

70%

70.2%

#43

Claude 3.5 Haiku Anthropic

69.2%

69%

69.2%

#44

Llama 3.1 70B Meta

68%

#45

Gemini 1.5 Pro Google

67.7%

68%

67.7%

#46

Claude 3 Opus Anthropic

60.1%

60%

60.1%

#47

Mixtral 8x22B Mistral AI

60%

#48

Llama 4 Scout Meta

50.3%

50%

50.3%

关于 MATH-500

全名: MATH Benchmark (500-problem subset)
类别: Math
指标: accuracy (%)
评分范围: 0%–100%
人类基准: 尚未确定
状态: 启用

Frequently Asked Questions

MATH-500 is a standardized evaluation that measures AI model performance on specific tasks. It provides comparable scores across different models, helping developers choose the right model for their needs.

o3 currently holds the top score on the MATH-500 benchmark. See our full rankings table above for the complete leaderboard with 48 models.

We update benchmark data from multiple sources including HuggingFace open-source model leaderboards and LMArena. Scores are refreshed regularly as new evaluations are published and new models are released.

No. While MATH-500 is an important indicator, real-world performance depends on many factors including pricing, latency, context window, and specific task requirements. We recommend using our composite score which weighs multiple benchmarks and practical factors.

MATH Benchmark (500-problem subset) 排行榜

模型排名

关于 MATH-500

相关基准测试