Qwen3 VL Instruct

Qwen3 VL Instruct is a vision-language mixture-of-experts model with 30 billion total parameters and 3 billion active parameters, designed for visual understanding, multimodal agent tasks, and high-resolution image analysis up to megapixel-level inputs. It supports strong multilingual OCR, visual grounding, GUI automation, and spatial reasoning — making it well-suited for document processing, visual question answering, and agentic workflows that require tight integration of language and vision.