We’ve rolled out some powerful updates to Document Parse (document-parse-250404) to make it even more flexible, accurate, and developer-friendly. From better form recognition to smoother handling of rotated and long document images — here’s what’s new this month.

1. Better table recognition in form-based documents

We’ve enhanced how Document Parse understands tables in complex form layouts. It now better handles multi-column formats, irregular structures, and layered sections — making it more accurate across real-world documents like invoices, registrations, and checklists.

2. Multi-page tables, stitched into one

Some documents include large tables that span multiple pages. Previously, these were treated as separate chunks — but with our new Layout Table Recognizer improvements, Document Parse can now detect and merge them into a single structured table.

3. Meet Document Parse 360°

Whether it’s 90°, -90°, or 180°, Document Parse now handles rotated inputs with ease. Even slightly tilted scans are no longer a problem. The result? Cleaner, more accurate parsing of any scanned or rotated document.

4. 2× faster processing for long images

We’ve optimized our patchify logic for long vertical documents — like claim forms or message logs. With better resizing, patch reduction, and GPU tuning, Document Parse now processes long images with significantly lower latency and memory usage.

Plus:

• Support for HWP conversion: You can now upload .hwp and .hwpx files — commonly used in Korea — and Document Parse will convert them automatically for processing.
• Smarter ocr=auto: Improved layout analysis automatically splits and structures complex documents with greater accuracy.

Curious how it all works? Try Document Parse yourself in the playground.
Contact us to share your challenges.

‍