Alibaba Group advanced its AI ambitions by releasing its most agentic-focused open-source code model to date.

The company stated Qwen3-Coder comes in multiple sizes, though highlighted a 450 billion parameter mixture-of-experts model with 35 billion active parameters offered outstanding performance on coding and agentic tasks.

Qwen3-Coder-480B-A35B-Instruct “sets new state-of-the-art results among open models” for agentic coding, and browser and tool use “compared to Claude Sonnet”.

Alibaba also highlighted “long-context capabilities” including “native support for 256K tokens, extendable up to 1 million tokens using Yarn”.

Agentic coding features are compatible with multiple platforms and offer various writing and instruction modes.

Alibaba released the Qwen3 large language model in April, with its AI moves coming at a time of growing focus on the technology among Chinese companies.

Cisco expects agentic AI to be widely used in various sectors by 2028.

Alibaba’s AI ambitions appear to be gaining traction, with a freshly-minted deal with HP announced on 22 July covering PCs and a recent high-ranking in generative AI numbers released by research company Omdia.