Compact models·EASYHUB JOURNAL
PaddleOCR-VL-1.6 keeps a compact architecture for complex documents
What changed
PaddleOCR-VL-1.6 retains a 0.9B architecture while improving tables, formulas, rare characters and difficult photographed or curved documents. The announcement’s 96.33 figure is tied to a specific OmniDocBench evaluation, not a universal accuracy guarantee. Representative business documents still require validation.
- Original title
- 百度文心发布 PaddleOCR-VL-1.6：准确率突破 96.33%，刷新文档解析 SOTA
- Source
- IT之家 · www.ithome.com
- Topic
- Compact models
- Source month
- 2026-06
This is a concise EasyHub summary of the linked source, not the full report or original reporting. Availability and preview conditions are described in the summary and original.
Summary page published · Editorial information