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.