Falcon OCR 1.5
Falcon OCR reads document images with a 270M parameter model. It uses one transformer for image patches and text. Version 1.5 replaced the repository's weights on 11 September 2026.
What changed in version 1.5
TII expanded training coverage for handwriting, historical scans, receipts and complex tables. It also applied GRPO reinforcement learning to reduce invented text and repeated rows. The category option selects the output format. A separate layout pipeline can divide a page before recognition.Falcon OCR model card and 1.5 update
Check structure as well as characters
Practical guidance. For an invoice workflow, compare the extracted cell positions and amounts with the source image. A readable transcript can still associate a value with the wrong row. Keep the page image beside the result when reviewing difficult scans.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | tiiuae/Falcon-OCRFalcon-OCR-1.5 |
| License | Apache 2.0Falcon OCR model card and 1.5 update |
| Outputs | Plain text, LaTeX formulas or HTML tablesFalcon OCR model card and 1.5 update |
| Repository | tiiuae/Falcon-OCR; the current card identifies version 1.5Falcon OCR model card and 1.5 update |
Related models
Falcon OCR Arabic adapts the 270M Falcon OCR architecture to Arabic documents.
Launch date not confirmedProvider product- Falcon Perception
Technology Innovation Institute (TII) · Falcon Perception
A 0.6B vision model that locates objects from text queries and returns pixel masks.
Launch date not confirmedDownloadable