Alibaba has announced the launch of Qwen3-Coder, its most advanced agentic AI coding model to date. Designed to revolutionize software development, Qwen3-Coder excels at complex agentic AI coding tasks, including generating new code, managing intricate coding workflows, and debugging across entire codebases.
Built on a Mixture-of-Experts (MoE) architecture, the open-sourced model, specifically Qwen3-Coder-480B-A35B-Instruct, features a total of 480 billion parameters, activating 35 billion parameters per token. This design prioritizes efficiency without compromising performance, allowing the model to achieve competitive results against leading state-of-the-art (SOTA) models across key benchmarks in agentic coding, browser use, and tool use.
In conjunction with the model, Alibaba is also open-sourcing Qwen Code, a powerful command-line interface (CLI) tool. This tool empowers developers to delegate engineering tasks to AI using natural language. Optimized with custom prompts and interaction protocols, Qwen Code is designed to unlock the full potential of Qwen3-Coder for real-world agentic programming. The model further supports integration with the Claude Code interface, simplifying coding tasks for developers.
Qwen3-Coder’s robust agentic coding capabilities stem from its training on an extensive dataset of codes and general text data. It natively supports a context window of 256K tokens, which can be extended up to 1 million tokens, allowing it to process vast codebases in a single session. Its superior performance is attributed to scaling across tokens, context length, and synthetic data during pre-training, as well as innovative post-training techniques like long-horizon reinforcement learning (agent RL).
This advancement enables the model to solve complex, real-world problems through multi-step interactions with external tools. As a result, Qwen3-Coder has achieved SOTA performance among open-source models on SWE-Bench Verified—a benchmark that evaluates AI models’ ability to solve real-world software issues—even without test-time or inference scaling.
Agentic AI coding is poised to transform software development by fostering more autonomous, efficient, and accessible programming workflows. With its open-source availability, robust agentic coding capabilities, and seamless compatibility with popular developer tools and interfaces, Qwen3-Coder is positioned as a valuable tool for global developers in software development.
The Qwen3-Coder-480B-A35B-Instruct model is now accessible on Hugging Face and GitHub. Developers can also utilize the model on Qwen Chat or through cost-effective APIs via Model Studio, Alibaba’s generative AI development platform.
Building on the success of previous iterations, Qwen-based coding models have already surpassed 20 million downloads globally. Tongyi Lingma, Alibaba Cloud’s Qwen-powered coding assistant, is slated for an upgrade, incorporating Qwen3-Coder’s enhanced agentic capabilities. Since its launch in June 2024, Tongyi Lingma’s “AI Programmer” feature—which offers code completion, optimization, debugging support, snippet search, and batch unit test generation—has generated over 3 billion lines of code.
