Mistral OCR 4

Mistral OCR 4 is a document-understanding model that converts PDFs, scans, and images into structured, machine-readable output — not just a wall of extracted text, but text tied to its position on the page. Each element comes back with a bounding box, a block classification (heading, paragraph, table, figure, and so on), and a per-element confidence score, across 170 languages. That structure is what separates it from plain text OCR: downstream systems can reason about layout, pull specific fields, and flag low-confidence regions for review instead of guessing.

Released June 23, 2026, it sits at the front of what Mistral is positioning as an enterprise document pipeline. You can hit the raw OCR model through the API at $4 per 1,000 pages (dropping to roughly $2 per 1,000 with the batch discount), or use the higher-level Document AI product at $5 per 1,000 pages when you want the extraction-to-fields workflow handled for you. The pricing is aggressive for what is a genuinely structured output, and the batch tier makes large back-catalog digitization projects realistic to budget.

It's built for teams processing documents at volume — invoice and receipt extraction, contract and form parsing, insurance and legal intake, and feeding clean, positioned text into RAG systems where layout context actually matters. It's an API-first, developer-facing product, so it competes less with consumer scan apps and more with Google Document AI, AWS Textract, and Azure Document Intelligence.