Qwen 2.5 VL 72B Instruct

Qwen2.5 VL 72B Instruct is Alibaba's January 2025 vision-language model, upgrading from Qwen2 VL with dynamic-resolution vision processing and streamlined visual encoding via window attention. It excels at recognizing objects, reading text within images, parsing charts and diagrams, and understanding long videos. Native spatial localization in JSON, including bounding boxes and coordinates at actual image scale, enables precise visual grounding. A 32K context window supports document processing and extended video sequences with event localization. Its agentic capabilities span computer-vision tasks, form extraction, and reasoning over embedded text, combining a dense 72B model with multimodal efficiency for production visual AI.