Alibaba has introduced Qwen3‑Coder, its most advanced open‑source AI coding model yet, using a 480 billion parameter Mixture‑of‑Experts architecture with 35 billion active parameters per token. It excels at autonomous “agentic AI coding tasks,” automating code generation, debugging, and complex workflow management. The model supports extremely long context windows—256 K tokens natively, extendable to 1 million.
Alibaba asserts it outperforms domestic rivals such as DeepSeek and Moonshot AI while matching benchmark results of GPT‑4 and Claude. It includes the open‑source CLI tool Qwen Code, adapted from Gemini Code, which enables prompt-based function calling and integrates smoothly with standard developer pipelines. Available via Hugging Face, GitHub, and Alibaba’s Model Studio under Apache 2.0 license, Qwen3‑Coder is positioned as a global open alternative for efficient, scalable software engineering.