CHURRO-DS
Stanford University
Historical document transcription dataset with 99,491 pages across 46 languages.
8
Total Results
6
Models Tested
2
Metrics
2025-12-21
Last Updated
Handwritten Score
Normalized Levenshtein Similarity on handwritten documents
Higher is better
| Rank | Model | Score | Source |
|---|---|---|---|
| 1 | churro-3b Historical handwritten documents, 46 languages, 99K pages | 70.1 | alphaxiv-leaderboard |
| 2 | gemini-25-pro | 63.6 | alphaxiv-leaderboard |
| 3 | gemini-25-flash | 58.7 | alphaxiv-leaderboard |
| 4 | qwen25-vl-72b | 54.5 | alphaxiv-leaderboard |
| 5 | claude-sonnet-4 | 37.1 | alphaxiv-leaderboard |
| 6 | gpt-4o | 34.2 | alphaxiv-leaderboard |
Printed Score
Normalized Levenshtein Similarity on printed documents
Higher is better
| Rank | Model | Score | Source |
|---|---|---|---|
| 1 | churro-3b Historical printed documents | 82.3 | alphaxiv-leaderboard |
| 2 | gemini-25-pro | 80.9 | alphaxiv-leaderboard |