Alibaba's Qwen3-Coder-Next Delivers Solid Coding Performance in Compact Software

2026-02-05

Alibaba has unveiled Qwen3-Coder-Next, a new open-source AI model designed for core programming tasks and local development. Trained on 800,000 verifiable tasks, the model boasts a total of 80 billion parameters, with only 3 billion active at any given time. Despite its relatively compact size, Alibaba claims it outperforms or matches larger open-source models on coding benchmarks, achieving scores above 70% on tests like SWE-Bench Verified and within the SWE-Agent framework. As always, benchmark results are indicative rather than definitive measures of real-world performance.

The model supports a context window of 256,000 tokens and is compatible with development environments such as Claude Code, Qwen Code, Qoder, Kilo, Trae, and Cline. It is also supported by local tools including Ollama, LMStudio, MLX-LM, llama.cpp, and KTransformers. Qwen3-Coder-Next is available under the Apache 2.0 license on platforms like Hugging Face and ModelScope.

RECENT AI NEWS

RECENT AI TOOLS