Datalab released OmniParseBench, an open benchmark for optical character recognition. It offers 16,288 yes/no tests that evaluate how well OCR parsers handle real documents. The tests cover 2,937 pages written in 92 different languages. Scores are reported as simple pass rates, and users can break them down by language, script or tag. The benchmark is freely available for anyone building or comparing OCR systems.
Why it matters
It lets developers and researchers compare OCR tools on a wide multilingual set, revealing strengths and gaps.
