Researchers introduced OvisOCR3, a 0.8-billion-parameter model designed for five OCR-related tasks, including document parsing, key information extraction and cross-page table reconstruction. The report says it scored 96.74 on OmniDocBench v1.6 and ranked first on the English and Chinese splits of OCRBench v2, with scores of 65.9 and 61.3. These are the researchers’ benchmark results; the report does not provide independent validation or specify the model’s access terms.
Why it matters
For teams handling documents, one model could cover several OCR tasks, but its real-world performance and access terms remain unclear.
