Document-intelligence OCR model that extracts structured, markdown-ready text from PDFs, slides, and images.
Last verified · 2026-06-23 · by Moe Ameen
Mistral OCR 4 is an optical character recognition and document-understanding model from Mistral, the French AI lab. It was released on June 23, 2026 as `mistral-ocr-4-0`, and Mistral's `mistral-ocr-latest` alias now points to it. Where a basic OCR engine hands you a flat wall of text, OCR 4 reads a document and returns a structured, machine-readable version of it.
The structure is the point. OCR 4 returns bounding boxes that localize each element on the page, typed-block classification that labels titles, tables, equations, signatures and other elements, and confidence scores reported per page and per word. The extracted text is formatted as clean markdown, which makes it usable directly for semantic chunking, RAG pipelines, and agent workflows. Mistral also positions OCR 4 as an ingestion component of its open-source Search Toolkit, feeding citation-ready inputs into retrieval and search workflows.
It handles PDF, DOC, PPT, and OpenDocument formats along with images, and supports 170 languages across 10 language groups, with Mistral reporting gains on rare and low-resource languages where several competing systems degrade. On public benchmarks the company reports 85.20 on OlmOCRBench and 93.07 on OmniDocBench, and says independent annotators preferred OCR 4 over competing systems at an average 72% win rate across 600+ multilingual documents. Treat specific scores as a snapshot and verify against the official source.
One thing to be clear about: OCR 4 is an extraction model, not a content generator. It reads documents and gives you structured text. It does not write social posts, draft a blog, or make video or images. Turning what it extracts into published content is a separate job.
Mistral OCR 4 solves the front-door problem: it turns a locked PDF, a scanned report, a slide deck, or a photo of a whiteboard into clean, structured text you can actually work with. What it does not do is turn that text into anything publishable. That handoff is where Kompozy comes in. Take the markdown OCR 4 returns, drop it into Kompozy as a source, and the engine fans one document into a full content set — a carousel that walks through the report's key findings, a LinkedIn post and an X thread written in your own voice through your Persona Brief, a blog article, an email newsletter, and a short script for a persona or avatar video — then schedules and publishes the set across all nine connected platforms from one queue.
The leverage is in the asymmetry. A 40-page whitepaper or a chart-heavy deck is hours of reading nobody on social will do. OCR 4 gets the words out in seconds; Kompozy turns those words into the week's posts. Because OCR 4 preserves tables and document structure as markdown, the source you hand Kompozy is clean enough that the generated carousel and blog track the actual document instead of drifting around a messy copy-paste.
Mistral OCR 4 is an OCR and document-understanding model from Mistral, released June 23, 2026 as mistral-ocr-4-0. It extracts structured, markdown-formatted text from documents and images, returning bounding boxes, typed-block classification (titles, tables, equations, signatures), and per-page and per-word confidence scores across 170 languages.
Mistral lists the API at $4 per 1,000 pages standard and $2 per 1,000 pages via the Batch API (a 50% discount), plus its Document AI product at $5 per 1,000 pages. A self-hosted option is available to enterprise customers. Check the official pricing page for current rates.
It handles PDF, DOC, PPT, and OpenDocument formats along with images, and supports 170 languages across 10 language groups, with reported gains on rare and low-resource languages.
No. OCR 4 extracts text from documents; it does not generate content or media. To turn the extracted text into posts, carousels, blogs, newsletters, or video, pass it to a content engine like Kompozy, which generates each format in your voice and publishes across platforms.
OCR 4 is the successor to Mistral OCR 3. Mistral reports higher benchmark scores and stronger multilingual and low-resource handling, along with structured output (bounding boxes, typed blocks, confidence scores) and a single-container self-hosting option. Verify specifics against Mistral's announcement.