Atlas

Benchmarks

← All benchmarks

LiveCodeBench - December 2023 to June 2024

Code

LiveCodeBench code-generation problems in the December 1, 2023 to June 1, 2024 date window (226 problems: 82 easy, 87 medium, 57 hard) used in the DeepSeek-Coder-V2 report (labelled 1201-0601). This exact historical problem cohort is retained as a diagnostic rather than pooled with other LiveCodeBench windows.

Top models (higher is better)

ModelScore
GPT-4 Turbo45.7
GPT-4o (2024-05-13)43.4
DeepSeek-Coder-V2-Instruct43.4
GPT-4 Turbo (1106 Preview)37.1
Opus 334.6
Codestral 240531.0
Llama 3 70B Instruct28.7
DeepSeek-Coder-V2-Lite-Instruct24.3
DeepSeek-Coder-Instruct 33B22.5
Loading Atlas data…