Alibaba unveils Qwen3-Coder, its most powerful open-source AI model for coding

Alibaba has taken another major step in the AI race with the launch of Qwen3-Coder, an open-source coding model it describes as its most advanced yet. The announcement was made on Wednesday, with the company highlighting the model’s strong performance in complex software development and agentic coding tasks.

Built on a Mixture-of-Experts architecture with 480 billion parameters and 35 billion active at a time, Qwen3-Coder supports a native context window of 256,000 tokens, which can scale up to 1 million tokens through extrapolation. According to a post on the Qwen X account, this version, officially titled “Qwen3-Coder-480B-A35B-Instruct”,  is Alibaba’s most powerful open agentic code model to date.

Qwen3-Coder is designed to tackle a wide range of developer tasks — from generating code to managing complex workflows autonomously. Alibaba claims that the model delivers top-tier results across multiple coding benchmarks, including SWE-bench-Verified, outperforming many other open models in this space.

To support developers and streamline integration, Alibaba is also releasing Qwen Code, a command-line tool forked from Gemini Code. This utility includes custom prompts and function-call protocols, optimised to unlock the full capabilities of Qwen3-Coder. According to the company, the tool is fully compatible with popular developer tools and ecosystems.