Qwen3-Coder: Powerful New AI Model Advances State-of-the-Art in Agentic Coding and Real-World Task Execution
Today, Alibaba Cloud announced the launch of Qwen3-Coder, its most advanced and agentic code model to date. Qwen3-Coder is available in various sizes, with the most powerful variant being Qwen3-Coder-480B-A35B-Instruct. This model boasts 480 billion parameters and 35 billion active parameters, supporting a context length of 256,000 tokens natively and 1 million tokens with extrapolation methods. Qwen3-Coder has set new benchmarks among open-source models in agentic coding, browser use, and tool interaction, comparable to the performance of Claude Sonnet 4. Alibaba Cloud is also open-sourcing a command-line interface (CLI) tool called Qwen Code. Adapted from Gemini Code, Qwen Code uses customized prompts and function calling protocols to unlock the full potential of Qwen3-Coder on agentic coding tasks. The tool integrates seamlessly with popular developer tools and is designed to enhance productivity in real-world software engineering scenarios. Pre-Training and Post-Training Enhancements To develop Qwen3-Coder, the team focused on scaling multiple dimensions during pre-training. This includes advancements in reinforcement learning (RL) for coding tasks, which are often characterized as "hard to solve, easy to verify." By scaling the RL training on a diverse set of real-world coding problems, the team automated the creation of high-quality training instances, significantly improving code execution success rates and enhancing performance across various tasks. Another crucial aspect of the post-training phase involved scaling long-horizon RL, essential for handling multi-turn interactions in software engineering tasks. This required overcoming the challenge of environment scaling. The team built a scalable system capable of running 20,000 independent environments in parallel, utilizing Alibaba Cloud's robust infrastructure. This system provided the necessary feedback for large-scale reinforcement learning, contributing to Qwen3-Coder's state-of-the-art performance on SWE-Bench Verified, a benchmark for real-world software engineering tasks. Using Qwen3-Coder Qwen Code To get started with Qwen Code, developers need to ensure they have Node.js 20+ installed. Installation can be done via the npm package manager: Install Node.js 20+ from the official repository. Install Qwen Code using npm: npm install -g qwen-code Alternatively, developers can install Qwen Code from the source: Clone the Qwen Code repository. Build and install the tool locally: git clone https://github.com/alibaba/qwen-code.git cd qwen-code npm install npm link Once installed, you can begin coding with Qwen Code by simply typing qwen in your terminal. Claude Code Integration Qwen3-Coder is also compatible with Claude Code, another popular coding tool. To use Claude Code with Qwen3-Coder: Request an API key from Alibaba Cloud Model Studio. Install Claude Code: npm install -g claude-code Configure the API with the claude-code-config npm package: npm install -g claude-code-config claude-code-config --api-key <your-api-key> --model qwen3-coder-plus --base-url https://dashscope-intl.aliyuncs.com/compatible-mode/v1 Start using Claude Code with Qwen3-Coder: ccr Use Cases Developers can leverage Qwen3-Coder through the Alibaba Cloud Model Studio API for a variety of tasks. This includes creating and testing code, interacting with browser-based tools, and integrating with custom software environments. Future Developments Alibaba Cloud is committed to further enhancing the capabilities of Qwen3-Coder. The team aims to expand its ability to tackle more complex and repetitive software engineering tasks, potentially liberating human developers from mundane work. They are also exploring possibilities for self-improvement in the Coding Agent—a groundbreaking concept that could revolutionize software development. Industry Insights and Company Profile The introduction of Qwen3-Coder underscores Alibaba Cloud's strategic focus on AI and its commitment to fostering innovation in the tech community. With Meta's significant investment in Scale AI, it’s clear that major players are prioritizing the development of advanced AI models to stay competitive. Qwen3-Coder, designed to integrate seamlessly with existing workflows, is expected to make a significant impact in the field of agentic coding. Industry experts predict that such advancements will not only improve the efficiency of software development but also open new avenues for automation and AI-assisted programming. Alibaba Cloud, a leading provider of cloud computing services, continues to push the boundaries of AI research and development. Their latest model, Qwen3-Coder, is part of a broader strategy to develop AI technologies that can solve real-world problems and enhance productivity across various sectors.