Character Error Rate leaderboard for optical character recognition systems evaluated on a mixed corpus of scanned documents, receipts, printed pages, and multilingual samples. Lower CER means fewer substituted, inserted, or deleted characters relative to ground truth.
Levenshtein-distance-based character error rate; lower is better.
Lower is better
Muted rows were not state of the art when published — an earlier or same-year result already scored better.
| Rank | Model | Trust | Score | Year | Links | Fix |
|---|---|---|---|---|---|---|
| 01 | mistral-ocr-3 | paper | 3.70 | 2025 | Source ↗ | Looks wrong? |