Mistral's new OCR model beats competitors in 72 percent of blind test cases, company says

2026-06-24 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Intermediate, quick

Summary

Mistral AI has released OCR 4, a new optical character recognition model designed to extract text from various document types, including PDFs, Word files, and PowerPoint presentations. Unlike previous versions, OCR 4 not only extracts raw text but also identifies the spatial location and semantic role of each element, classifying them as titles, tables, equations, or signatures. This block classification feature aids in automatically segmenting documents for search systems or AI agent processing. The model provides confidence scores for words and pages, indicating its certainty. OCR 4 supports 170 languages, including less common ones. In a blind test involving over 600 documents, independent reviewers preferred OCR 4's results 72 percent of the time compared to competing models. It is accessible via API, Mistral Studio, and Microsoft Foundry, costing \$4 per 1,000 pages, or \$2 in batch mode.

Key takeaway

For AI Engineers or Product Managers integrating OCR solutions, Mistral's OCR 4 offers advanced capabilities beyond raw text extraction. You should consider its semantic block classification and confidence scores for improved document processing and AI agent workflows. Its reported 72 percent blind test preference and 170-language support suggest a robust option for diverse, complex document types. Evaluate its \$4 per 1,000 pages cost against your project's specific needs.

Key insights

Mistral's OCR 4 extracts text and semantically classifies document elements, outperforming competitors in blind tests.

Principles

• Semantic classification enhances OCR utility.
• Confidence scores improve reliability assessment.
• Blind testing validates real-world performance.

Method

OCR 4 identifies text elements, their page location, and semantic role (title, table, equation, signature), then outputs confidence scores for words and pages.

In practice

• Feed classified blocks into search systems.
• Enable AI agents to process structured documents.
• Utilize 170-language support for diverse content.

Topics

• OCR 4
• Mistral AI
• Optical Character Recognition
• Document Processing
• Semantic Classification
• Multilingual Support
• AI Agents

Best for: AI Architect, NLP Engineer, CTO, AI Engineer, Machine Learning Engineer, AI Product Manager

Related on AIssential

• OCR Processor - Mistral AI — Mistral AI's OCR API extracts structured text from diverse document types, preserving formatting and providing confidence scores.
• Mistral Document AI (with OCR 4) and Mistral Medium 3.5 arrive in Microsoft Foundry — Mistral's new models in Microsoft Foundry enable integrated, structured document processing and advanced reasoning for enterprise AI.
• DeepSeek OCR Review — DeepSeek OCR offers rapid, accurate document-to-markdown conversion, outperforming vision LLMs in speed and data fidelity.
• Deep Dive into Semantic Formatting Score: A New Metric for Meaningful Document Formatting — Document OCR benchmarks must evaluate semantic formatting for accurate AI agent interpretation.
• Mistral drops Voxtral Transcribe 2, an open-source speech model that runs on-device for pennies — Mistral's Voxtral Transcribe 2 offers accurate, cost-effective, on-device speech-to-text with strong privacy features.
• [R] IDP Leaderboard: Open benchmark for document AI across 16 VLMs, 9,000+ documents, 3 benchmark suites — An open benchmark evaluates 16 VLMs on document understanding, revealing tight performance margins among top models.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.