Qwen3 VL 235B A22B Thinking

Qwen3 VL 235B A22B Thinking combines 235 billion parameters (about 22B activated) with explicit step-by-step reasoning for complex multimodal problems. The thinking mode generates reasoning before the final answer, exposing how the model interprets images, video, and text. With a 131K context window, it handles video analysis, multi-page documents, and spatial reasoning where understanding the visual reasoning process is useful. The model unifies strong text generation with visual perception and is tuned for multimodal reasoning in STEM and math. It fits visual reasoning benchmarks, complex document analysis that benefits from step-by-step verification, and research applications where interpretable multimodal reasoning is required. The open-weight release enables fine-tuning on specialized visual reasoning domains.