Alibaba Enters the Vibe Coding Scene with Qwen 3 Coder

Alibaba team has officially launched Qwen3-Coder, its most advanced open-source coding model yet. The flagship variant—480B-parameter Mixture-of-Experts model with 35B active parameters—now supports an impressive 256K context window, expandable to 1 million tokens, revolutionising large-scale code understanding and agent workflows.

Qwen3-Coder excels at agentic coding—autonomous programming tasks—delivering performance on benchmarks like SWE-bench Verified on par with proprietary systems such as Claude Sonnet. The model natively supports browser and tool use, real-time function calling, and runs without separate thinking prompts.

"Qwen3-Coder-480B-A35B-Instruct sets new state-of-the-art results among open models on Agentic Coding, Agentic Browser-Use, and Agentic Tool-Use, comparable to Claude Sonnet 4," the company said in a blog post.

Alongside the model, Alibaba open-sourced Qwen Code, a command-line interface forked from Gemini CLI. It’s tailored for agent interactions and integrates seamlessly with community tools, enabling developers to build agentic workflows efficiently.

The architecture benefits from advanced training: Qwen3-Coder was trained on 7.5T tokens (70% code) using large-scale reinforcement learning and long-horizon optimization. Infrastructure supports 20,000 parallel environments to improve tool and code execution learning, boosting real-world accuracy.

With native 256K context and up to 1M token support, Qwen3-Coder is optimized for repository-scale tasks, personal coding assistants, and multilanguage programming.

Available now via GitHub, Hugging Face, and conversational APIs, it positions itself as a top-tier open-source rival in agentic AI coding.