Mistral OCR 4 Self-Hosting: What Mistral Confirms
What Mistral confirms about self-hosting OCR 4, keeping document data in your environment, deployment options, pricing and practical limits.
Tags
Quick summary
What Mistral confirms about self-hosting OCR 4, keeping document data in your environment, deployment options, pricing and practical limits.
Mistral OCR 4 Self-Hosting: What Mistral Confirms
Mistral AI announced OCR 4 on June 23, 2026 as a document-intelligence model that extracts structured content. For organisations evaluating private document processing, the important distinction is between the public API, Document AI and Mistral’s self-hosting option. This article confines itself to what Mistral’s announcement confirms.
What self-hosting means in the announcement
Mistral describes OCR 4 as compact enough to deploy in a single container. It says a fully self-hosted deployment can keep document data inside an organisation’s environment for residency, sovereignty and compliance needs. The announcement also says that self-managed deployment is available to enterprise customers.
That is not the same as a consumer desktop installation guide. The announcement does not publish a hardware matrix, a downloadable model-weight package, an Ollama model identifier or a command-line setup procedure. Teams should therefore treat any such instructions as unverified unless Mistral publishes a specific deployment document for them.
What OCR 4 returns
OCR 4 returns extracted text together with bounding boxes, typed block classification and inline confidence scores. Mistral gives titles, tables, equations and signatures as examples of block types. Its announcement states support for 170 languages across 10 language groups and lists PDF, DOC, PPT and OpenDocument among supported enterprise formats.
Those outputs can help a self-hosted document pipeline retain the text, its location, its type and a confidence signal. Mistral positions the model for document ingestion, enterprise search, RAG and structured extraction workflows.
API, Document AI and self-hosting
Mistral says OCR 4 is available through one API endpoint. Document AI uses the same OCR engine and adds parameters for schema-shaped JSON, image annotation and custom instructions. The announcement lists API pricing at $4 per 1,000 pages; the stated Batch API discount produces $2 per 1,000 pages; Document AI is listed at $5 per 1,000 pages.
Self-hosting is the relevant option when a qualified enterprise needs document data to remain in its own environment. API and Document AI are relevant when a team needs managed integration and the announced service model fits its requirements. A deployment decision should account for privacy, volume, evaluation results and the organisation’s operational responsibilities rather than assuming that every OCR 4 use is offline or local by default.
Limits to keep in view
Mistral calls OCR 4 a document-understanding model, not a decision-maker. It says the model is not intended for medical diagnosis, legal advice or judgment, high-stakes financial decisions, safety-critical systems, real-time or latency-sensitive processing, or non-document inputs such as raw audio and video.
Mistral also cautions that aggregate benchmark results are directional and recommends evaluation on representative documents. A self-hosting decision should therefore include a test on the documents, accuracy needs, latency constraints and compliance requirements that actually apply.
