Most organizations are sitting on a mountain of documents they cannot fully use. Contracts, invoices, technical reports, and compliance filings pile up in formats that are hard to search and harder to analyze. The first wave of AI tools helped with simple text extraction, but converting a scanned page to raw text is only the first step. What happens next, figuring out which part is a table, which is a heading, and which is a signature, still required custom code or a human.
Mistral OCR 4 is designed to close that gap. Released on June 23, it reads documents and returns structured output: every piece of content is tagged by type, located with a coordinate on the page, and accompanied by a confidence score at the word level.
The practical effect is significant. A system receiving this output does not just know what the document says. It knows where the invoice total sits, how certain the extraction was, and whether a human reviewer needs to check that section. That is the difference between a tool that extracts and one that organizes.
The pricing is worth taking seriously. At $4 per 1,000 pages via the standard API, and $2 per 1,000 pages in bulk, the cost of processing large document archives drops to a level that makes projects previously considered impractical worth attempting. One customer reported equivalent accuracy at roughly eight times lower cost and seventeen times lower latency compared to competing tools for financial document work. Another reported speeds four times faster than their previous provider.
The document processing market itself is large and growing fast. The global market for intelligent document processing was valued at around $2.3 billion in 2024 and is projected to reach $12.35 billion by 2030. Banking, insurance, and financial services currently account for the largest share of that spend, with healthcare growing fastest.
Where Mistral's timing is interesting is the competitive field. Until recently, the market was split between legacy providers like ABBYY and Kofax, which built on-premises products sold through enterprise licensing deals, and cloud services from Google, Amazon, and Microsoft. Mistral enters with a model that competes on accuracy, undercuts on price, and offers the self-hosted option those legacy players are known for.
For organizations in regulated industries, that last point matters most. Healthcare providers, insurers, financial firms, and public sector organizations in many countries face rules that prevent them from sending sensitive documents to third-party cloud services. OCR 4 can run inside their own infrastructure on a single standard server container, which means the tool is accessible to them in a way that Google Document AI or Azure OCR often is not.
The 170-language support also opens doors beyond English-speaking markets. The model performs well on rare and lower-resource languages where competing systems typically degrade. For multinational operations, or for companies in markets where European, Arabic, or Asian language documents dominate, this is a practical advantage.
The benchmark numbers should be read with appropriate skepticism. Mistral itself flags limitations in the testing methodology, including potential errors in ground-truth data and reading-order assumptions that can penalize correct outputs. The company also does not fully disclose the methodology behind the 72 percent win rate in human-preference tests, which makes independent verification difficult.
Where the story is clearest is at the pricing and deployment level. If your business regularly processes large volumes of documents, and most do, the combination of structured output, per-word confidence scoring, and a low per-page cost gives teams a practical reason to revisit what they have been building and buying in this space.