Qwen Team Releases Qwen3-Coder, a Large Agentic Coding Model with Open Tooling

Qwen Team released Qwen3-Coder, a family of AI models specifically designed for complex, multi-step programming tasks. The flagship 480B parameter model uses a Mixture-of-Experts architecture and supports up to 1 million tokens for repository-scale code analysis. Unlike traditional code generators, it emphasizes execution and decision-making through reinforcement learning training on real-world tasks. The release includes Qwen Code, an open-source CLI tool for experimentation, and API access through DashScope. While powerful, the larger models require significant infrastructure for local deployment.