Mistral, an artificial intelligence company, released a new version of its document-reading software on 23 June 2026, called OCR 4, announced on its website.

The software is designed to read and understand documents the way humans do. Instead of simply converting scanned text into characters — which is what older scanning software does — Mistral's tool also understands the structure of documents. It recognizes where tables are, reads equations, identifies images, and understands how all these pieces relate to each other on the page.

Mistral first released this technology in March 2025 and immediately built it into Le Chat, its consumer AI assistant. Le Chat is used by millions of people, so the software gets tested on real documents every day. OCR 4 benefits from this accumulated experience.

Document-reading accuracy has become important for modern AI systems. Many companies now use AI to process large numbers of documents automatically. If the document reader fails — breaking up tables, skipping equations, or mixing up multi-column layouts — the AI that relies on that information makes mistakes. A good document reader is therefore essential, not optional. Mistral's decision to release a fourth version of its OCR software shows the company believes this capability matters.

One thing worth noting is Mistral's strategy of making OCR 4 the default tool in Le Chat, rather than selling it only as a standalone service. By processing millions of real documents from millions of people, Mistral gets feedback that helps it spot and fix problems that don't show up in controlled tests. This real-world experience is an advantage that purely API-based competitors don't have.

If you work in a company that processes documents automatically, the release of OCR 4 means Mistral is committed to improving this tool. Full details about how the model performs and what it can do are in the announcement.