Qwen2.5-Coder Series: Powerful, Diverse, Practical.

GitHub Pages · 1 year ago · 23

Alibaba open-sourced the Qwen2.5-Coder series, a collection of code generation models ranging from 0.5B to 32B parameters designed for code assistance and artifact creation. The flagship 32B model achieved state-of-the-art performance on multiple benchmarks including EvalPlus, LiveCodeBench, and BigCodeBench, matching GPT-4o's coding capabilities and scoring 73.7 on the Aider code repair benchmark. The release provides developers with six model sizes under permissive licenses (mostly Apache 2.0) to integrate into code assistants like Cursor and creative applications that generate websites and visualizations.