Alibaba Group announced on Wednesday the launch of Qwen3-Coder, an open-source AI model for software development, claiming it to be its most advanced coding tool to date.
The launch comes amid intensifying competition among Chinese technology companies in the global AI development race, with firms on both sides of the Pacific releasing increasingly sophisticated models.
Forked from Gemini Code, Qwen Code has been adapted with customised prompts and function calling protocols to unleash the full capabilities of Qwen3-Coder on agentic coding tasks. Qwen3-Coder works seamlessly with the community’s best developer tools.
“As a foundation model, we hope it can be used anywhere across the digital world, Agentic Coding in the World,” Alibaba said in a statement.
To address scaling challenges, the company has developed a system that runs 20,000 independent environments in parallel on Alibaba Cloud. This enables Qwen3-Coder to achieve state-of-the-art performance on SWE-Bench Verified without requiring test-time scaling.
Based on performance data published by Alibaba, Qwen3-Coder surpassed domestic rivals, including models from DeepSeek and K2 by Moonshot AI, in essential coding abilities. The company also asserted that its model performed on par with top US models, such as Anthropic’s Claude and OpenAI’s GPT-4, in specific areas.
“Unlike the prevailing focus on competitive-level code generation in the community, we believe all code tasks are naturally well-suited for execution-driven large-scale reinforcement learning. That’s why we scaled up Code RL training on a broader set of real-world coding tasks,” the company said.
By automatically increasing the variety of coding task test cases, Alibaba developed high-quality training instances, effectively harnessing the full potential of reinforcement learning.
The company claims that this method has improved code execution success rates and fostered exploration of challenging, easily verifiable tasks as promising opportunities in large-scale reinforcement learning.