Qwen3.5 makes multimodality part of the foundation

The Qwen team introduced Qwen3.5, emphasizing native vision-language processing and agentic capabilities. The announcement describes a mixture-of-experts architecture, an expanded language range, and both open-model and hosted offerings. Its design treats visual information as part of the model’s foundation.

Context

Users should distinguish the downloadable model from the hosted service: context limits, tools, and deployment conditions may differ. A broad language count also does not guarantee equal quality in every language. The relevant test combines the user’s actual inputs, required tools, and expected output format.