
Qwen3.5
The 397B native multimodal agent with 17B active params
Qwen is a native vision‑language model that supports long‑horizon agentic workflows. Its hybrid design combines linear attention with a Mixture‑of‑Experts structure, providing intelligence comparable to a 397 billion‑parameter system while running at the speed of a 17 billion‑parameter model. The model offers deep multimodal reasoning, large context windows and efficient execution for complex real‑world tasks.