OCRBench
multimodal
+
+
+
+
About
OCRBench is a comprehensive optical character recognition benchmark for evaluating text recognition capabilities of multimodal models. It tests models' ability to accurately recognize, extract, and understand text from images across various formats, fonts, languages, and visual contexts, providing systematic assessment of OCR performance and text-centric visual understanding.
+
+
+
+
Evaluation Stats
Total Models7
Organizations3
Verified Results0
Self-Reported7
+
+
+
+
Benchmark Details
Max Score1
Language
en
+
+
+
+
Performance Overview
Score distribution and top performers
Score Distribution
7 models
Top Score
88.5%
Average Score
84.6%
High Performers (80%+)
7Top Organizations
#1Alibaba Cloud / Qwen Team
3 models
87.5%
#2Microsoft
1 model
84.4%
#3DeepSeek
3 models
81.8%
+
+
+
+
Leaderboard
7 models ranked by performance on OCRBench
| License | Links | ||||
|---|---|---|---|---|---|
| Jan 26, 2025 | tongyi-qianwen | 88.5% | |||
| Aug 29, 2024 | tongyi-qianwen | 87.7% | |||
| Jan 26, 2025 | Apache 2.0 | 86.4% | |||
| Feb 1, 2025 | MIT | 84.4% | |||
| Dec 13, 2024 | deepseek | 83.4% | |||
| Dec 13, 2024 | deepseek | 81.1% | |||
| Dec 13, 2024 | deepseek | 80.9% |
