Qwen2-VL: dynamic resolution, computer use, and doc-level OCR at 72B

Qwen2-VL is an Alibaba model capable of analyzing images of any size without cropping or resizing — it processes them at native resolution. It can read entire documents, perform OCR on PDF pages, and even control a computer by watching the screen as a human would. With 72 billion parameters it is among the most powerful open multimodal models ever released.