OmniDocBench
OmniDocBench is a benchmark for document parsing: it scores how well a model turns PDF pages of many types into text, tables and formulas.
OmniDocBench was introduced in a December 2024 paper, with code and data on GitHub under opendatalab. Its pages come from nine document sources, including academic papers, textbooks, handwritten notes and dense newspapers, with annotations for 19 layout categories and 15 attribute labels. Most OCR model cards on this site quote a score on it.
The benchmark has been revised, with versions such as v1.5 and v1.6, and derived sets such as Real5-OmniDocBench, which the PaddlePaddle team built from the v1.5 data with scanned, warped and photographed pages. Scores from different versions are not comparable, and the numbers on this site are the developers' own. Run a model on your own documents to see how it copes.
Related terms
- OCR and document parsing OCR turns an image of text into text; document parsing goes further and rebuilds tables, formulas, charts and reading order, usually as Markdown.
- Benchmark A benchmark is a standardized test used to compare models, such as a set of maths, coding or knowledge questions; scores are useful but easy to over-read.