What is Mistral OCR?
Mistral OCR is the document-reading API behind Mistral AI's Document AI service. You send a PDF, Word, PowerPoint or image file to the /v1/ocr endpoint and get back Markdown for each page, tables as Markdown or HTML, extracted images, headers and footers, and, on the current model, paragraph-level bounding boxes, 13 types of block label and confidence scores at page, block or word level. The same endpoint can also return JSON that matches a schema you supply.
The mistral-ocr-latest alias points to OCR 4.1, released on 16 July 2026 and generally available since 31 August; OCR 4.0 was retired on 30 September. Our model review of OCR 4.1 covers its benchmarks in detail.
It costs $4 per 1,000 pages, $2 through the Batch API, and $5 per 1,000 pages when you ask for schema-matched annotations. It is also sold through Amazon SageMaker and Microsoft Foundry, and as a single self-hosted container for enterprise customers.
How much does Mistral OCR cost?
| Plan | Price | What you get |
|---|---|---|
| OCR API | $4 per 1,000 pages | Markdown, tables, bounding boxes, block labels, confidence scores |
| Batch API | $2 per 1,000 pages | |
| Annotations | $5 per 1,000 pages | JSON matching your schema |
| Self-hosted container | Custom | Enterprise customers |
$4 per 1,000 pages; $2 per 1,000 pages on the Batch API; $5 per 1,000 annotated pages. Self-hosted container for enterprise customers, price not published.
Is Mistral OCR worth it?
You have a folder of scanned contracts in French, Polish and Arabic, and the tool you tried turned every table into a jumble of words.
Mistral OCR reads each page and gives you clean Markdown, with tables kept as tables and images pulled out separately. The current model, OCR 4.1, also marks where each paragraph sits on the page, labels it as a title, table, equation, signature or one of nine other types, and scores how confident it is. That last part matters more than it sounds: instead of trusting every page, you send only the uncertain ones to a person.
It covers 170 languages. Mistral reports 85.20 on OlmOCRBench and 93.07 on OmniDocBench, but those are its own figures, and it calls its comparisons with competitors directional.
The catch: the price went up. At $4 per 1,000 pages it costs more than double Azure Document Intelligence's $1.50 for plain text, and the older Mistral OCR rate of $1 that many guides still quote no longer applies. The Batch API halves it to $2, which is the right setting for anything that is not needed in seconds. Files are capped at 50 MB and 1,000 pages. On the free API tier, your documents may be used for training unless you opt out.
Our take: for a European company that wants layout-aware reading with EU hosting, or an on-premises container, this is the natural choice, and on batch it undercuts Reducto's $10 parse by a wide margin. For the same five invoice formats every month, Azure's prebuilt models do the job for less.
Where it earns its place
A legal-tech startup indexing EU court filings in 20 languages
Make scanned judgments searchable with links back to the exact paragraph
Bounding boxes tie each extracted paragraph to its place on the page for citations. A 200,000-page backlog costs about $400 on the Batch API, against $800 at the standard rate.
An operations analyst at a mid-sized logistics firm
Turn customs forms into rows in a database without writing a parser
Send each form with a JSON schema and get matching fields back at $5 per 1,000 annotated pages; route low-confidence blocks to a person for checking.
A head of IT at a German bank
Run OCR on statements without documents leaving the bank's own servers
Mistral offers OCR as a single self-hosted container for enterprise customers. Pricing and terms are by arrangement and not published.
Before you commit
- Setup
- Create a Mistral API key and call /v1/ocr with model mistral-ocr-latest from the Python or TypeScript SDK or curl; pass a public URL, a Base64 file or an uploaded file.
- Integrations
- Mistral API and Batch API, Document AI annotations, Amazon SageMaker and Microsoft Foundry; inputs include PDF, DOCX, PPTX, PNG, JPEG and AVIF.
- Data and privacy
- Mistral's help centre says free-tier data may be used for training unless you opt out, and pay-as-you-go users can opt out; a self-hosted container keeps documents on your own infrastructure for enterprise customers.
- Cost at scale
- One million pages costs $4,000 a month on the standard API, $2,000 on batch, or $5,000 with annotations.
- Limits
- Files must be 50 MB or less and 1,000 pages or fewer; block labels and confidence need OCR 4 or newer, and Mistral says it is not meant for high-stakes medical, legal or financial decisions.
What's good
- Paragraph bounding boxes, block labels and confidence scores
- 170 languages
- Batch API halves the price to $2 per 1,000 pages
- Schema-matched JSON from the same endpoint
- Self-hosted container for enterprise customers
What's not
- $4 per 1,000 pages, four times the old $1 rate
- Benchmark comparisons are Mistral's own
- 50 MB and 1,000-page limit per file
- Free-tier documents may be used for training unless you opt out
Best for
Developers building document pipelines in many languages who need page locations and confidence scores, especially in Europe or on-premises.
Mistral OCR alternatives
Mistral OCR: common questions
- Is Mistral OCR worth it?
- Mistral OCR reads documents well in almost any language and tells you how sure it is, at a price that only makes sense once you switch on batch. We rate it 4.2 out of 5.
- How much does Mistral OCR cost?
- $4 per 1,000 pages; $2 per 1,000 pages on the Batch API; $5 per 1,000 annotated pages. Self-hosted container for enterprise customers, price not published.
- What are the downsides of Mistral OCR?
- $4 per 1,000 pages, four times the old $1 rate; benchmark comparisons are Mistral's own; 50 MB and 1,000-page limit per file; free-tier documents may be used for training unless you opt out.
- Who is Mistral OCR best for?
- Developers building document pipelines in many languages who need page locations and confidence scores, especially in Europe or on-premises.
- How much does Mistral OCR cost for a team or in production?
- One million pages costs $4,000 a month on the standard API, $2,000 on batch, or $5,000 with annotations.
- Is Mistral OCR safe for company data?
- Mistral's help centre says free-tier data may be used for training unless you opt out, and pay-as-you-go users can opt out; a self-hosted container keeps documents on your own infrastructure for enterprise customers.
- What are the best Mistral OCR alternatives?
- The closest alternatives we have reviewed are Reducto, LlamaParse and Azure Document Intelligence.
Answers checked Oct 26, 2026.
Make Mistral OCR? Show this rating on your site
