Qwen3-Coder: Agentic Coding in the World
9 months ago
- #AI
- #Machine Learning
- #Coding
- Announcement of Qwen3-Coder-480B-A35B-Instruct, a 480B-parameter Mixture-of-Experts model with 35B active parameters.
- Supports context length of 256K tokens natively and up to 1M tokens with extrapolation methods.
- Sets new state-of-the-art results in Agentic Coding, Agentic Browser-Use, and Agentic Tool-Use.
- Open-sourcing Qwen Code, a command-line tool for agentic coding, adapted from Gemini Code.
- Pre-training advancements include scaling tokens (7.5T tokens with 70% code ratio), context, and synthetic data.
- Post-training focuses on Code RL and Long-Horizon RL for real-world coding tasks and multi-turn interactions.
- Installation and usage instructions for Qwen Code and Claude Code provided.
- API access through Alibaba Cloud Model Studio demonstrated with a Python example.
- Future work includes improving Coding Agent performance, releasing more model sizes, and exploring self-improvement.