Document parsing sub-score from the IDP Leaderboard, based on OmniDocBench evaluating text extraction, formula recognition, table structure, and reading order across 1,355 pages and 18,000+ samples in Chinese and English.
Comprehensive parsing evaluation; covers 9 document types; bilingual; uses established metrics (Edit Distance, TEDS, CDM).
Composite score averages diverse sub-tasks; a model strong on text may be weak on formulas.
Higher is better. Composite parsing score, 0-100.
Max score: 100
Composite score averages diverse sub-tasks.
| # | Model | Score |
|---|---|---|
| 1 | 90.1 | |
| 2 | 90.1 | |
| 3 | 90.0 | |
| 4 | 89.5 | |
| 5 | 88.8 | |
| 6 | 88.0 | |
| 7 | 86.9 | |
| 8 | 85.9 | |
| 9 | 85.3 | |
| 10 | 85.3 | |
| 11 | 85.3 | |
| 12 | 82.5 | |
| 13 | 81.9 | |
| 14 | 79.9 | |
| 15 | 79.6 | |
| 16 | 79.6 | |
| 17 | 78.3 | |
| 18 | 76.7 | |
| 19 | 76.4 | |
| 20 | 69.2 | |
| 21 | 67.6 | |
| 22 | 63.4 | |
| 23 | 59.7 | |
| 24 | 48.7 | |
| 25 | 47.3 | |
| 26 | 43.3 | |
| 27 | 42.3 |