Qwen3.5 is a reasoning vision-language model that supports tool use. With 35B total parameters and 3B activated, it outperforms previous generation models more than 6x its size.

31B dense Gemma4 variant. Supports vision input, reasoning, and tool calls.