Mistral OCR 4: Features, Pricing and Limits
Announced June 23, 2026, Mistral OCR 4 extracts structured document content with bounding boxes, block types and confidence scores.
Tags
Quick summary
Announced June 23, 2026, Mistral OCR 4 extracts structured document content with bounding boxes, block types and confidence scores.
Mistral OCR 4: Features, Pricing and Limits
Mistral AI announced Mistral OCR 4 on June 23, 2026. It is a document-intelligence model for extracting and structuring document content. This article summarizes facts confirmed by Mistral and does not present that announcement as current news.
Structured document output
Mistral says OCR 4 returns extracted text, bounding boxes, typed block classification and inline confidence scores. Its examples of block types include titles, tables, equations and signatures. An application can therefore retain text together with the position, type and confidence signal of a detected element.
The announcement states support for 170 languages across 10 language groups. It lists PDF, DOC, PPT and OpenDocument among supported enterprise document formats.
Deployment and integration
Mistral describes OCR 4 as compact enough to run in a single container. It presents fully self-hosted deployment for data-residency, sovereignty and compliance needs, and says self-managed deployment is available to enterprise customers.
The API is intended for extraction workflows. Mistral also describes Document AI as an application-level option built on the same OCR engine, with schema-shaped JSON and image annotation options when the relevant parameters are used.
Prices stated by Mistral
The announcement lists API pricing at $4 per thousand pages. A 50% Batch API discount reduces that price to $2 per thousand pages. Document AI is listed at $5 per thousand pages. Recheck official pricing before a production integration.
Uses and limits
Mistral positions OCR 4 for document ingestion, enterprise search, RAG, archive digitisation and structured extraction such as invoice fields. It also says OCR 4 is a document-understanding model, not a decision-maker. It is not intended for medical diagnosis, legal advice or judgment, high-stakes financial decisions, safety-critical systems, real-time or latency-sensitive processing, or raw audio and video.
Mistral publishes benchmark and human-evaluation results but warns that aggregate benchmark scores are directional. Evaluate the model on representative documents and the required accuracy, latency, privacy and cost constraints.
