Alibaba's open model Qwen3.6 leads Google's Gemma 4 across agentic coding benchmarks

Alibaba launched Qwen3.6-35B-A3B, an open mixture-of-experts model that uses only 3 of its 35B parameters at a time.

The company says it outperforms its Qwen3.5 predecessor and Google’s Gemma 4 on coding, reasoning, and multimodal benchmarks.

Alibaba also says the model matches Claude Sonnet 4.5 on some image and video tasks.

Qwen3.6-35B-A3B is available in Qwen Studio, Alibaba Cloud Model Studio, and for download on Hugging Face and ModelScope.