Newsbite analysis · AI, Tools
The Decoder
Mistral's new OCR model beats competitors in 72 percent of blind test cases, company says
Mistral AI has unveiled its OCR 4 model, which processes text from various document formats while classifying content elements such as titles and tables. This innovation has achieved a notable 72% preference rate in blind tests against competitors, showcasing its potential for improved document handling and AI integration. The model supports 170 languages, making it versatile in application.
Act nowConfidence: HighShort term horizon
Published 24 Jun 2026, 10:10 UTC1 min read
Lead image — The Decoder
Brief
NewsBite reading:Mistral AI launches OCR 4, outperforming rivals in 72% of tests
TL;DR
- Mistral's OCR 4 model reads from PDFs, Word files, and presentations.
- It identifies content roles, enhancing document meaningfulness.
- The model offers confidence scores for its text readings.
- OCR 4 supports 170 languages, including many less common ones.
- In testing, reviewers preferred OCR 4's results 72% of the time.
Interesting facts
- The model's ability to classify document elements revolutionizes how text data can be processed.
- Mistral's pricing model is competitive, encouraging wider adoption.
What changed
The introduction of OCR 4 marks a significant advancement in text recognition technology.