Qwen3-Coder is the Qwen team's family of open coding models, built to act as agents that read, write and test code across whole projects.
In plain words
The first and largest release, from July 2025, is a mixture-of-experts model with 480 billion parameters, 35 billion active at a time. It reads 256,000 tokens natively, enough for a large code repository. It was pre-trained on 7.5 trillion tokens, mostly code, then trained with reinforcement learning on tasks whose results can be checked by running tests. Smaller versions followed, and the team also released a terminal coding tool, Qwen Code.
A quick example
A developer could point a coding tool such as Qwen Code or Cline at a Qwen3-Coder model and ask it to add a feature. The model reads the relevant files, writes changes and runs the tests to check them.
Where you'll meet it
- Agentic coding in terminal tools and IDE plugins.
- Code completion and fill-in-the-middle editing.
Related
Sources
- DocsQwen3-Coder: Agentic Coding in the World, Qwen Team · read 29 Sept 2026
- DocsQwen3-Coder, QwenLM (GitHub) · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.