How do I know if my OCR is good enough?
Test against the same benchmarks that state-of-the-art models use. 26 standard tests across documents, handwriting, video, scene text, and multilingual tasks.
Top Models Across Multiple Benchmarks
Models ranked by number of top-3 finishes across different benchmarks.
gemini-25-pro
gpt-4o
paddleocr-vl
mineru-2.5
claude-sonnet-4
gemini-15-pro
qwen25-vl-72b
qwen25-vl-32b
dots-ocr-3b
mistral-ocr-2512
Document (9)
OHRBench
No data yetShanghai AI Laboratory
LLMDoc
No data yetAlibaba Group
Misraj-DocOCR
No data yetMisraj
EPHOIE
No data yetSouth China University of Technology
Multilingual (6)
KRETA
No data yetSeoul National University
KOCRBench
No data yetKL-Net
MOTBench
No data yetShanghai Jiao Tong University
JaPOC
No data yetFast Accounting Co., Ltd.
Scene Text (5)
DROBS
No data yetNVIDIA
CORU
No data yetUniversity of Innsbruck
Union14M
No data yetSouth China University of Technology
TextZoom
No data yetNanjing University
ICDAR 2015
No data yetMegvii Inc.
Video (3)
VTPBench
No data yetSouth China University of Technology
VideoDB OCR
No data yetVideoDB
Handwriting (3)
UIT-HWDB
No data yetUniversity of Information Technology
Chinese Text Recognition
No data yetFudan University
Digital Peter
No data yetMoscow Institute of Physics
About This Directory
This directory tracks 26 OCR benchmarks. Benchmarks with green badges have full leaderboard data - click them to see all models and their scores.
Data sourced from AlphaXiv. Last updated: 2025-12-21.
Related Reading
dots.ocr 3B
88.41 OmniDocBench, 100+ languages
Mistral OCR 3
79.75 composite, $2/1000 pages
clearOCR
Traditional OCR, Polish service
Model Comparator
Interactive side-by-side comparison
Decision Guide
Which OCR for your use case?
Vendor Comparison
Pricing, accuracy, deployment options