Mistral OCR 4 Document Intelligence Model
Mistral released OCR 4 on 23 June 2026, describing it as breakthrough document parsing that returns bounding boxes, block classification and inline confidence scores alongside extracted text, supports 170 languages and runs in a single container for fully self-hosted deployments.
Tool Overview
Features, steps and FAQ below
Features
- ✓ Mistral says OCR 4 returns a structured representation of the document rather than just clean text: each block is localized with a bounding box, classified by type (titles, tables, equations, signatures and more), with per-page and per-word confidence scores
- ✓ Mistral says it supports 170 languages across 10 language groups, including specialized and low-resource languages many systems handle poorly, and accepts common enterprise formats including PDF, DOC, PPT and OpenDocument
- ✓ Mistral says OCR 4 is compact enough to deploy on a single container for fully self-hosted deployments, keeping document data inside an organization's own infrastructure for residency, sovereignty and compliance, while supporting high-throughput batch processing
- ✓ Mistral says independent annotators preferred OCR 4 over every leading OCR and document-AI system tested with win rates averaging 72%, and it scored the top overall result on OlmOCRBench (85.20); OCR 4 is an ingestion component of Mistral's open-source Search Toolkit (public preview)
- ✓ Mistral says OCR 4 through the API is priced at $4 per 1,000 pages with a 50% Batch-API discount (reducing it to $2 per 1,000 pages), while Document AI is $5 per 1,000 pages, with a self-hosting option available
How to Use
- Call the OCR 4 model via the API through Mistral Studio, or use Document AI, an application-level, no-code path to the same engine
- Follow the official Getting Started with OCR 4 Cookbook for a first extraction and to work with bounding boxes and block classification
- For strict data-privacy, sovereignty or compliance requirements, choose a single-container self-hosted deployment so sensitive documents stay within your own infrastructure
- When you need structured output, add the Document AI parameters to the same call (pass a JSON schema or a custom prompt) to reshape OCR output to your spec or annotate detected images
FAQ
What is Mistral OCR 4?
Mistral says OCR 4 is its document intelligence model released on 23 June 2026, returning bounding boxes, block classification and inline confidence scores alongside extracted text, and supporting 170 languages.
How is OCR 4 different from ordinary OCR?
Mistral says where previous generations focused on converting a page into clean text and tables, OCR 4 returns a structured representation of the document: each block is localized with a bounding box, classified by type, with per-page and per-word confidence scores, supporting source-grounded citations, redactions and human-in-the-loop verification.
Which languages and file formats does OCR 4 support?
Mistral says it supports 170 languages across 10 language groups, including specialized and low-resource languages, and accepts common enterprise formats including PDF, DOC, PPT and OpenDocument.
How much does OCR 4 cost?
Mistral says OCR 4 through the API is priced at $4 per 1,000 pages, with a 50% Batch-API discount reducing it to $2 per 1,000 pages, while Document AI costs $5 per 1,000 pages.
Can I self-host OCR 4?
Mistral says yes: OCR 4 is compact enough to deploy on a single container for fully self-hosted deployments, keeping document data within your own infrastructure for residency, sovereignty and compliance.