MISTRAL DROPS OCR 4 AND BETS ENTERPRISE CLIENTS WILL PAY TO KEEP DOCUMENTS OFF THE CLOUD The dirty secret of enterprise AI adoption is that most companies cannot actually use the big cloud AI services for their most important documents. Contracts, patient records, financial filings, internal legal memos — these are exactly the documents that would benefit most from AI processing, and they are exactly the documents that compliance departments, data privacy laws, and common sense say should not leave the building. Mistral is betting that problem is worth solving. OCR 4 is a document intelligence model that can run in a single container on a company’s own hardware. It extracts structured data from documents across 170 languages, classifies document blocks, includes bounding boxes and confidence scores alongside extracted text, and delivers citation-ready outputs directly into retrieval-augmented generation pipelines. Independent testing puts it first on OlmOCRBench with a score of 85.20, with annotators preferring it over competing systems 72 percent of the time. It is available through Mistral’s API for $4 per 1,000 pages, or $2 per 1,000 pages via batch processing. Enterprise customers who want to run it internally get full self-hosted deployment. It also connects to Amazon SageMaker and Microsoft Foundry. Mistral is targeting one billion euros in revenue for 2026. OCR 4 is the kind of enterprise product that makes that number believable. Keywords: Mistral OCR 4, document AI enterprise, self-hosted AI document extraction, Mistral enterprise AI