OCRBench — leaderboard
OCR and document understanding benchmark: 1000 question-answer pairs testing text recognition, scene text VQA, document VQA, key information extraction, and handwritten math expression recognition.
Metric: Final Score. Source: huggingface.co. 29 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Minicpm-V 2.6 | 852 |
| 2 | granite-vision-3.3-2b | 824 |
| 3 | MiniMonkey | 806 |
| 4 | H2OVL-Mississippi-2B | 782 |
| 5 | InternVL2-1B | 779 |
| 6 | InternVL2-4B | 776 |
| 7 | InternVL2-2B | 768 |
| 8 | H2OVL-Mississippi-0.8B | 751 |
| 9 | Qwen-VL-Max | 723 |
| 10 | Qwen-VL-Plus | 694 |
Interactive version: theaggregate.ai/benchmark?slug=ocrbench · How the rankings work · Data refreshed daily, snapshot 2026-07-22.