Posted by Hangzhou, China – Alibaba has unveiled its most advanced agentic AI coding model yet, called Qwen3-Coder. Built to deliver high-performance software development, the model can perform a range of tasks including, generating new code, managing complex coding workflows and debugging entire codebases.
Built on a Mixture-of-Experts architecture and open source under the name Qwen3-Coder-480B-A35B-Instruct, Qwen3-Coder comprises 480 billion parameters in total but activates 35 billion parameters per token, offering a balance of efficiency and strong performance. The model has achieved competitive results against leading state-of-the-art models across key benchmarks in agentic coding, browser usage and tool utilisation.
Alibaba has also open-sourced Qwen Code, a powerful command-line interface tool that allows developers to delegate engineering tasks to the AI using natural language. Optimised with custom prompts and interaction protocols, Qwen Code enables users to fully leverage Qwen3-Coder’s capabilities in real-world programming. Furthermore, the model supports integration with the Claude Code interface to make coding tasks even more accessible to developers.
Qwen3-Coder model leveraged an extensive dataset, comprising both code and general text data for training, and engineered for robust agentic coding. It supports a native context window of 256,000 tokens and can be extended up to one million tokens, allowing it to process large codebases in a single session. Its impressive performance comes from a combination of scaling across tokens, longer context lengths, synthetic data during pre-training, and innovative post-training methods such as long-horizon reinforcement learning.
This enables Qwen3-Coder to solve complex real-world problems through multi-step interactions with external tools. As a result, it achieves state-of-the-art performance among open-source models on the SWE-Bench Verified benchmark, which evaluates AI models’ ability to address real software issues, even without additional scaling during inference.
Agentic AI coding is transforming software development by enabling more autonomous, efficient and accessible workflows. Given its open-source availability, strong agentic coding abilities, and seamless compatibility with popular developer tools and interfaces, developers around the world consider Qwen3-Coder as a valuable resource.
The Qwen3-Coder-480B-A35B-Instruct model is now available on Hugging Face and GitHub. Developers can also access the model on Qwen Chat or via cost-effective APIs through Model Studio, Alibaba’s generative AI development platform.
Qwen-based coding models have already surpassed 20 million downloads globally. Alibaba Cloud’s Qwen-powered coding assistant Tongyi Lingma will soon be upgraded with Qwen3-Coder’s enhanced agentic features.
Since its launch in June 2024, Tongyi Lingma’s AI Programmer feature, which offers code completion, optimisation, debugging support, snippet search and batch unit test generation, has produced over three billion lines of code.