What’s the story
Alibaba has unveiled Qwen3-Coder, its most sophisticated open-source artificial intelligence (AI) model for software development.
The company said the new model is capable of handling complex coding tasks and workflows, thereby improving code generation and management efficiency with agentic AI technology.
The launch comes amid fierce competition among Chinese tech companies in the global AI development race.
Qwen3-Coder outperforms several popular models
Alibaba has positioned Qwen3-Coder as a model particularly strong in “agentic AI coding tasks.” These are automated processes where AI systems can independently tackle programming challenges.
Performance data released by Alibaba shows that the new model outperformed domestic competitors such as DeepSeek and Moonshot AI’s K2 in key coding capabilities.
It also matched the performance of leading US models, including Anthropic’s Claude and OpenAI’s GPT-4, in certain areas.
What is the Qwen3-Coder model?
The Qwen3-Coder model is a 480-billion parameter Mixture-of-Experts (MoE) model with 35 billion active parameters. It natively supports a context of up to 256,000 tokens and can scale up to one million tokens through extrapolation.
Despite being available in different sizes, Alibaba has introduced the ‘most powerful’ version yet.
The company also open-sourced a command-line tool for agentic coding called Qwen Code along with the new model.
Alibaba is working on improving the coding agent
Alibaba is actively working to improve the performance of its coding agent with Qwen3-Coder. The goal is to handle more complex and tedious tasks in software engineering, thus enhancing human productivity.
The company has also teased more model sizes of Qwen3-Coder that will deliver strong performance while cutting down on deployment costs.