Benchmarks
Three complementary benchmarks evaluate document AI models across different dimensions of intelligent document processing.
OlmOCR Bench
v1.0OCR unit tests on 1,403 pages. Math, tables, reading order, headers, watermarks.
Metrics
MathTablePresentAbsentOrder
18 models evaluated·1,403 pages · 7,010 tests
Leading model
Nanonets OCR2+ 82.2%
OmniDocBench
v1.5Full page parsing on 1,355 pages. Text, formulas, tables, reading order.
Metrics
Text Edit↓CDM↑TEDS↑TEDS-S↑Read Order↓
18 models evaluated·1,355 pages
Leading model
Gemini-3-Flash 90.1%
IDP Core Bench
v1.0Production doc tasks on ~2,000 documents. KIE, OCR, tables, VQA.
Metrics
KIEOCRTableVQA
17 models evaluated·~2,000 documents
Leading model
Gemini 3.1 Pro 89.6%