BENCHMARKS Open benchmarks, every parser we could run, every test public. ← All benchmarks
OVERALL · OMNIPARSEBENCH

Overall: Datalab accurate scores 93.5% on OmniParseBench.

OmniParseBench is our open parsing benchmark: 16,288 pass/fail tests on 2,937 pages in 92 languages. The headline is the mean of three family scores (text, tables and layout), and an open-code scorer checks every test with no model judging the outputs.

SCOREBOARD · OmniParseBench

Ranked scoreboard.

Models ranked highest-to-lowest. Datalab variants in accent; competitors and prior generations in muted ink.

Rank Model Score vs scale
01 Datalab accurate current model, Oct 20 93.5%
02 Gemini 3.8 Flash* 92.9%
03 Datalab balanced 91.6%
04 Reducto 90.8%
05 LlamaParse 90.4%
06 Mistral OCR 89.5%
07 Claude Sonnet 5.5* 88.7%
08 GPT-5.6 Sol 87.8%
09 Extend 83.3%
10 Azure 83.0%
Dataset · OmniParseBench Last run · 2026-10-08

+2.7 vs the next vendor API

16,288 tests · 2,937 pages · 92 languages

* Used in creating and verifying at least some OmniParseBench tests.

WHY THIS NUMBER IS CREDIBLE

Every OmniParseBench test is a yes/no question about one page, and every score is the share of tests passed. The scorer is open code with no model judging the outputs. Each tag score here is the mean of the text, tables and layout pass rates on the tests with that tag. Score any slice yourself in the OmniParseBench post, or download the tests from the Hugging Face dataset.

START

Try our hosted API on your hardest PDF.

Free tier, no credit card. Every OmniParseBench test and verdict is public.