Qwen3.6-27B beats much larger predecessor on most coding benchmarks

Alibaba has released Qwen3.6-27B, a new dense open-source model with 27 billion parameters. According to Alibaba, the model outperforms its much larger predecessor, Qwen3.5-397B-A17B (397 billion parameters) on nearly every coding benchmark tested. It scored 77.2 on SWE-bench Verified compared to 76.2, and 59.3 on Terminal-Bench 2.0 compared to 52.5.

The model handles both text and multimodal reasoning. As a "dense" model, it's easier to run than the more complex MoE (Mixture of Experts) architectures, which activate different sub-models depending on the task.

Qwen3.6-27B is available through Qwen Studio, the Alibaba Cloud Model Studio API, and as open weights on Hugging Face and ModelScope. It's aimed at developers who want strong coding performance without dealing with a massive model.

As always, benchmark results only hint at real-world performance, and efficient Chinese open-source models might benefit from research and development coming out of Western AI labs.