Atlas

Benchmarks

← All benchmarks

LiveCodeBench - September 2023 to April 2024

Code

LiveCodeBench code-generation problems in the September 1, 2023 to April 1, 2024 date window that DeepSeek used for the DeepSeek-V2 chat comparison (labelled 0901-0401 on the DeepSeek-V2 model card). This exact historical problem cohort is retained as a diagnostic rather than pooled with other LiveCodeBench windows.

Top models (higher is better)

ModelScore
DeepSeek-V2-Chat32.5
Llama 3 70B Instruct30.5
Mixtral 8x22B Instruct25.0
Qwen1.5 72B Chat18.8
DeepSeek LLM 67B Chat18.3
Loading Atlas data…