Search & fact check with cited sources for What are the key features, benchmark performance, pricing options, and strategic context of Mistral AI's newly launched OCR 4 model for document processing?

Here is the fact-checked breakdown of Mistral AI's newly released OCR 4 model.

Key Features

• Structure-aware extraction — Returns paragraph-level bounding boxes, block-type labels (title, text, table, equation, signature, etc.), and per-word confidence scores alongside extracted text, not just plain text dumps .
• include_blocks API parameter — When enabled, each page returns a blocks array with structural labels and spatial coordinates .
• 170 languages across 10 language groups, with particular accuracy gains on rarer/less-resourced scripts .
• Single-container self-hosting — The model can be deployed entirely on-premises in one container for organizations that need data sovereignty .
• Multimodal input — Accepts PDFs and images (Office documents via conversion), outputs structured Markdown and JSON, designed as an ingestion component for RAG and agentic pipelines .

Benchmark Performance

Benchmark	OCR 4 Score	Notes
OlmOCRBench (public leaderboard)	85.20	Topped the leaderboard at launch
OmniDocBench	93.07	Strong across diverse document types
Human preference (blind, 600+ docs, 12+ languages)	72% average win rate	Independent annotators preferred OCR 4 over competing OCR systems

Mistral also reports strong scores on its internal Crawl Multilingual benchmark, though the raw numbers were not published in the sources reviewed .

Pricing Options

Tier	Price	Details
Standard OCR	$4 per 1,000 pages	Base text extraction
Annotated (structured)	$5 per 1,000 pages	Includes bounding boxes, block labels, and confidence scores

Pricing is page-based, not token-based, which is unusual among Mistral's other models and reflects the document-batch use case.

Strategic Context

• Launched June 23–24, 2026 alongside Mistral Document AI (the end-to-end pipeline product that wraps OCR 4) and Mistral Medium 3.5, all arriving simultaneously in Microsoft Azure AI Foundry .
• Shift from "text extraction" to "document understanding" — OCR 4 is positioned as a fundamental layer for enterprise search, RAG pipelines, and agentic workflows where preserving layout and structure (tables, equations, signatures) is critical .
• Competitive positioning — Directly targets Google's Document AI, Azure Document Intelligence, and open-source OCR pipelines by combining structured output at commodity pricing with a self-hostable container option, which is rare among major OCR APIs .
• Single-container design is a key differentiator for regulated industries (finance, healthcare, legal) that cannot send documents to external APIs .