最近更新: 2h ago

Arena 基准测试

LiveBench (Dynamic) 排行榜

Comprehensive benchmark across 6 categories (math, coding, reasoning, data analysis, instruction following, language) using contamination-resistant, regularly updated questions.

为什么重要： Contamination-free by design — uses new questions regularly. Top models still score below 70%, making it highly discriminating.

顶级模型

87.3%

o4 Mini High

平均评分

64.7%

共51个模型

已测试模型

指标: average score

人类基准

评分范围: 0%–100%

LiveBench Scores - Top 25 Models

Ranked by LiveBench score (%)

LMMarketCap.com

模型排名

All models with a reported LiveBench score, ranked by highest average score.

排名模型评分性能

o4 Mini High OpenAI

87.3%

87%

87.3%

R1 0528 DeepSeek

84.4%

84%

84.4%

Qwen3 235B A22B Alibaba

80.4%

80%

80.4%

Claude 3.5 Sonnet Anthropic

80%

Gemini 2.0 Flash Google

80%

Gemma 4 31B Google

80%

GPT-5.4 OpenAI

79%

Claude Opus 4.6 Anthropic

78.5%

79%

78.5%

Gemini 3 Pro Google

78%

#10

GPT-5.2 OpenAI

77.5%

78%

77.5%

#11

GPT-5 OpenAI

76%

#12

Gemini 2.5 Pro Google

75.6%

76%

75.6%

#13

o3 OpenAI

74.8%

75%

74.8%

#14

Gemini 3 Flash Google

74.5%

75%

74.5%

#15

Claude Opus 4.5 Anthropic

74%

#15

GPT-5.1 OpenAI

74%

#17

Claude Sonnet 4.6 Anthropic

73.2%

73%

73.2%

#18

Grok 4 xAI

73%

#19

Claude Opus 4 Anthropic

72%

#19

o4-mini OpenAI

72%

#21

o3-mini OpenAI

71.5%

72%

71.5%

#22

Claude Sonnet 4.5 Anthropic

70.5%

71%

70.5%

#23

DeepSeek R1-0528 DeepSeek

69.5%

70%

69.5%

#24

DeepSeek R1 DeepSeek

67.3%

67%

67.3%

#25

Claude Sonnet 4 Anthropic

67%

#26

Claude 3.5 Haiku Anthropic

66.7%

67%

66.7%

#26

Claude 3 Opus Anthropic

66.7%

67%

66.7%

#28

Gemini 2.5 Flash Google

66.5%

67%

66.5%

#29

Claude 3.7 Sonnet Anthropic

65.8%

66%

65.8%

#30

GPT-4.1 OpenAI

64.5%

65%

64.5%

#31

GPT-4o OpenAI

64.3%

64%

64.3%

#31

GPT-4o (2024-11-20)OpenAI

64.3%

64%

64.3%

#33

Grok 3 xAI

62.5%

63%

62.5%

#34

Llama 4 Maverick Meta

62%

#35

DeepSeek V3 DeepSeek

60.4%

60%

60.4%

#36

Qwen 2.5 Coder 32B Alibaba

57.1%

57%

57.1%

#36

Mistral Large 2 Mistral AI

57.1%

57%

57.1%

#36

Grok 2 xAI

57.1%

57%

57.1%

#39

Gemini 1.5 Pro Google

53.3%

53%

53.3%

#39

Llama 3.1 70B Meta

53.3%

53%

53.3%

#39

DeepSeek V3 (March 2025)DeepSeek

53.3%

53%

53.3%

#39

Command A Cohere

53.3%

53%

53.3%

#39

Gemini 2.0 Flash Lite Google

53.3%

53%

53.3%

#39

R1 Distill Llama 70B DeepSeek

53.3%

53%

53.3%

#45

GPT-4o-mini (2024-07-18)OpenAI

50%

#46

DeepSeek V3 DeepSeek

49.6%

50%

49.6%

#47

Command R (08-2024)Cohere

46.7%

47%

46.7%

#48

GPT-4 Turbo OpenAI

37.3%

37%

37.3%

#49

Phi-4 Microsoft

35.7%

36%

35.7%

#50

R1 Distill Qwen 32B DeepSeek

33.3%

33%

33.3%

#51

Claude 3 Haiku Anthropic

22.5%

23%

22.5%

关于 LiveBench

全名: LiveBench (Dynamic)
类别: Arena
指标: average score (%)
评分范围: 0%–100%
人类基准: 尚未确定
状态: 启用

Frequently Asked Questions

LiveBench 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.

o4 Mini High currently holds the top score on the LiveBench benchmark. See our full rankings table above for the complete leaderboard with 51 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 LiveBench 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.

LiveBench (Dynamic) 排行榜

模型排名

关于 LiveBench

相关基准测试

LiveBench (Dynamic) 排行榜

模型排名

关于 LiveBench

相关基准测试