Datalab has released OmniParseBench, an open OCR benchmark designed to audit how well document parsers handle optical character recognition. The benchmark includes 16,288 yes/no tests drawn from 2,937 pages spanning 92 languages, giving developers a broad set of checks for evaluating parser accuracy.
Each score in OmniParseBench is reported as a pass rate, and users can slice results by tag, script, or language. That granularity makes it possible to see where a parser struggles—for example, with a particular script or document type—rather than relying on a single aggregate number.
The benchmark is openly available, which means teams can run the same tests against their own OCR systems and compare results. Datalab positions OmniParseBench as a practical tool for auditing parser performance across a wide range of languages and page layouts.