Qwen2.5 VL! Qwen2.5 VL! Qwen2.5 VL!

Qwen's Qwen2.5-VL can return structured information about where content appears in an image. The release describes object locations as boxes or points and supports machine-readable coordinates and attributes. For documents, its QwenVL HTML format represents layout information as well as extracted content. The official account identifies uses involving scanned forms, invoices and tables.

This gives downstream software more than a prose description to work with: it can receive identified content together with location or layout information. The model represents object coordinates using the image's actual scale, making the relationship between its output and the source image explicit. The capability provides a foundation for document processing without establishing error-free extraction.