Atlas

Benchmarks

← All benchmarks

LiveCodeBench - May 2023 to September 2024

Code

LiveCodeBench code-generation problems in the May 2023 to September 2024 date window that Qwen labels "LiveCodeBench 2305-2409". This exact historical problem cohort is retained as a diagnostic rather than pooled with other LiveCodeBench windows.

Top models (higher is better)

ModelScore
Qwen2.5 72B Instruct55.5
Claude 3.5 Sonnet (June 2024)54.1
Qwen Plus (rolling alias)51.4
Qwen2.5 Instruct 32B51.2
DeepSeek-V2.549.4
Qwen-Turbo (rolling alias)43.1
Qwen2.5 14B Instruct42.6
Mistral Large 2 (Instruct 2407)42.2
Llama 3.1 405B Instruct41.6
GPT-4o Mini40.7
Qwen2.5 14B Instruct 1M38.6
Qwen2.5-Turbo37.8
Qwen2 72B Instruct32.2
Llama 3.1 70B Instruct32.1
Qwen2.5 7B Instruct28.7
Qwen2.5 7B Instruct 1M28.0
Qwen2 7B Instruct23.9
Qwen2-57B-A14B Instruct22.5
Qwen2.5 3B Instruct19.9
Gemma 2 9B IT18.9
Phi-3.5 Mini Instruct15.8
Qwen2.5 1.5B Instruct14.8
Llama 3.1 8B Instruct8.3
Gemma 2 2B IT5.8
Qwen2.5 0.5B Instruct5.1
Qwen2 1.5B Instruct4.5
Qwen2 0.5B Instruct1.6
Loading Atlas data…