PaddleOCR-VL-1.6
PaddleOCR-VL-1.6 is a 0.9B document vision-language model released in May 2026. It retains the 1.5 architecture while improving training for weak document regions.
Model design and access
The update uses region-aware data selection and progressive post-training. Its target cases include complex tables, rare characters, historical Chinese text, seals, text spotting and charts. The model weights are one part of the wider PaddleOCR parsing pipeline.PaddleOCR-VL-1.6 official model card
Practical use
Practical guidance. Compare the regions that fail in your current parser before changing the entire workflow. A model update may improve cell text while a separate layout or table-merge component still controls document structure. Evaluate both the model output and the assembled page result.PaddleOCR-VL-1.6 official model card
Specifications & access
| Detail | Value and conditions |
|---|---|
| Launch date | |
| Access | Downloadable |
| License | Apache 2.0PaddleOCR-VL-1.6 official model card |
| Parameters | 0.9 billionPaddleOCR-VL-1.6 official model card |
Related models
- PaddleOCR-VL-1.5
Baidu · PaddleOCR-VL
A 0.9-billion-parameter document model for skewed, warped or unevenly lit pages, with Tibetan and Bengali support.
Launch date not confirmedDownloadable - Qianfan-OCR
Baidu · Qianfan-OCR
Four-billion-parameter document model for recognition, layout analysis, question answering and field extraction in 192 languages.
Launch date not confirmedDownloadable