Mistral OCR

What Mistral OCR is

Mistral OCR is Mistral’s service for understanding documents and extracting their text. OCR 4.1 provides paragraph-level bounding boxes, structural labels for blocks and confidence scores at the block level. OCR 4 supports 170 languages across 10 language groups, along with table extraction, structure extraction and handwriting OCR. Its maximum file size is 512 MB.

Teams can submit documents through the API, including batches through the Batch API, or use Document AI in Mistral Studio as a no-code interface. OCR 4 can run in a single container on an organization’s infrastructure; self-managed deployment is available to enterprise customers. Mistral says self-hosting keeps document data in the customer’s environment for residency, sovereignty and compliance needs. The service is also an ingestion component of Mistral Search Toolkit, an open-source framework for RAG and enterprise search. Python and TypeScript SDKs are listed. The OCR 4.1 API plan is $4 a month, billed at $4 per 1,000 pages. Document AI is $5, paid once, billed at $5 per 1,000 annotated pages. Mistral says OCR is not intended for real-time processing, non-document inputs, medical diagnosis, legal advice or judgment, high-stakes financial decisions or safety-critical systems.

Who it suits

Mistral OCR suits teams that need to extract text and structure from documents, including handwriting and tables, through an API or a no-code interface. Organizations with data residency needs can consider self-hosting, while enterprise customers can use self-managed deployment and Enterprise API support with system-level SLAs.

Getting started

1. Visit the Mistral OCR website.
2. Choose the OCR 4.1 API or Document AI plan for the relevant processing option.
3. Use the API for document submissions, including batches through the Batch API.
4. Use Document AI in Mistral Studio for the no-code interface.
5. For self-hosting, run OCR 4 in a single container on your infrastructure; self-managed deployment is available to enterprise customers.

Verdict

Choose Mistral OCR for document text and structure extraction, with API batching, a no-code interface and self-hosted deployment options. Look elsewhere if you need real-time processing, non-document inputs or support for the high-stakes uses Mistral excludes.