Qwen3 VL 235B A22B Instruct

Qwen3 VL 235B A22B Instruct is the direct-answer checkpoint in the large Qwen3-VL MoE tier. It accepts text and images and connects visual content with text knowledge. Of its 235 billion total parameters, 22 billion are active. Instruct fine-tuning targets controllable answers rather than making visible Thinking the defining behavior.

The catalog lists 131,072 context tokens and up to 32,768 output tokens. For multi-step visual analysis, the separate Qwen 3 VL 235B A22B Thinking checkpoint is the relevant alternative. The two checkpoints have different post-training goals.