EuroDocBench
A synthetic benchmark of European invoice extraction: 12 locales, 5 layouts, born-digital PDFs and
simulated scans. Systems must return 12 fields in canonical form — converting "1 234,56 €" to 1234.56
and "5. března 2026" to 2026-03-05 is part of the task.
Code ·
Dataset ·
Report
Leaderboard (test split, 1,200 documents)
Browse documents
| Field | Gold | Prediction |
|---|
✓ correct · ✗ wrong · ∅ no value
All documents, companies, VAT numbers and IBANs are synthetic; identifiers deliberately fail their official checksums. Built by Since AI.