Qwen3.6 35B A3B

Native vision-language MoE model with hybrid linear attention and sparse mixture-of-experts, improved agentic coding, math/code reasoning, and spatial intelligence.