Large-scale bilingual text-centric benchmark with 4x more tasks than OCRBench v1, covering 31 diverse scenarios for text localization and reasoning in multimodal models.
Comprehensive OCR evaluation beyond simple text extraction; bilingual (Chinese/English); tests reasoning about text in images.
More research-oriented than enterprise-focused; includes scene text alongside documents.
Higher is better. Composite score across 31 OCR scenarios.
Max score: 100
More research-oriented than enterprise-focused.
| # | Model | Score |
|---|---|---|
| 1 | 92.4 | |
| 2 | 92.2 | |
| 3 | 91.1 | |
| 4 | 90.8 | |
| 5 | 90.5 | |
| 6 | 90.4 | |
| 7 | 89.8 | |
| 8 | 89.7 | |
| 9 | 89.5 | |
| 10 | 89.1 | |
| 11 | 89.0 | |
| 12 | 88.8 | |
| 13 | 88.2 | |
| 14 | 88.1 | |
| 15 | 88.1 | |
| 16 | 88.0 | |
| 17 | 87.9 | |
| 18 | 87.8 | |
| 19 | 87.5 | |
| 20 | 87.5 | |
| 21 | 87.5 | |
| 22 | 87.4 | |
| 23 | 87.3 | |
| 24 | 87.3 | |
| 25 | 87.0 | |
| 26 | 87.0 | |
| 27 | 87.0 | |
| 28 | 87.0 | |
| 29 | 86.9 | |
| 30 | 86.9 | |
| 31 | 86.9 | |
| 32 | 86.9 | |
| 33 | 86.8 | |
| 34 | 86.8 | |
| 35 | 86.8 | |
| 36 | 86.8 | |
| 37 | 86.8 | |
| 38 | 86.7 | |
| 39 | 86.7 | |
| 40 | 86.7 | |
| 41 | 86.6 | |
| 42 | 86.6 | |
| 43 | 86.5 | |
| 44 | 86.5 | |
| 45 | 86.5 | |
| 46 | 86.4 | |
| 47 | 86.4 | |
| 48 | 86.4 | |
| 49 | 86.3 | |
| 50 | 86.3 |