LiveCodeBench (2407-2501)Reasoning

LiveCodeBench was introduced in 2024 by Princeton and other institutions. It continuously collects new problems from LeetCode, AtCoder, and CodeForces for contamination-resistant code evaluation. It supports code generation, self-repair, code execution, and test-output prediction with pass@1 scoring; v6 contains 1,055 problems.

Ranking Sources

#ModelScore

1

Gemini 2.5 ProGoogle

80.1%

2

GLM-4.5Zhipu AI

72.9%

3

GLM-4.5-AirZhipu AI

70.7%

4

DeepSeek-R1-Distill-Llama-70BDeepSeek

57.5%

5

DeepSeek-R1-Distill-Qwen-32BDeepSeek

57.2%

6

OpenAI o1-miniOpenAI

53.8%

7

DeepSeek-R1-Distill-Qwen-14BDeepSeek

53.1%

8

QwQ-32B-PreviewAlibaba

41.9%

9

DeepSeek-R1-Distill-Llama-8BDeepSeek

39.6%

10

Claude 3.5 Sonnet 20241022Anthropic

38.9%

18 models total

Back to benchmarks