Measured in the open, against every parser we could run.
OmniParseBench is our open parsing benchmark: 16,288 pass/fail tests on 2,937 pages in 92 languages, scored by open code with no model as judge. Datalab accurate scores 94.1, the top score of every parser we ran.
94.1 accurate · 92.6 balanced. The top score of every parser we ran.
The headline is the mean of three family scores: text, tables and layout. The Datalab rows come from production runs on October 23, published in the benchmark's results file. Every other row is the leaderboard run in the OmniParseBench post. Pick any slice of the tests, like handwritten tables or Thai text, and score it yourself in the post.
- Reducto
- LlamaParse
- Mistral OCR
- Extend
- Azure
- Gemini 3.8 Flash*
- Claude Sonnet 5.5*
- GPT-5.6 Sol
- Datalab accurate
- Datalab balanced
* Used in creating and verifying at least some OmniParseBench tests. Tesseract ran too but has no headline, because it cannot output layout. 16,288 tests · 2,937 pages · 92 languages.
Scores on the documents you parse.
Pick the slice closest to your documents. Each one has a ranked table of every parser and the tests behind it.
Try our hosted API on your hardest PDF.
Free tier. No credit card. Capacity for 1B+ pages in a day.