End-to-end document parsing benchmark covering text recognition, table recognition, formula recognition, reading order, and layout detection across 9 document types (academic papers, textbooks, slides, financial reports, newspapers, handwritten notes) in Chinese and English.
Most comprehensive single benchmark for document parsing quality; CVPR 2025 standard; covers 1,729 pages across 19 layout categories.
Composite score averages across diverse tasks; a model good at text may score poorly on formulas; Chinese/English only.
Higher is better. Composite score across text, table, formula, reading order, and layout tasks.
Max score: 100
Composite score averages across diverse tasks.
| # | Model | Score |
|---|---|---|
| 1 | 235.0 | |
| 2 | 90.4 | |
| 3 | 90.2 | |
| 4 | 88.0 | |
| 5 | 87.0 | |
| 6 | 80.3 | |
| 7 | 75.0 | |
| 8 | 4.0 | |
| 9 | 3.0 | |
| 10 | 1.0 | |
| 11 | 0.9 |