What is Mistral OCR 4.1?
Mistral OCR 4.1 is Mistral AI's document-reading model, the engine behind its Document AI service. OCR 4 launched on 23 June 2026, OCR 4.1 followed on 16 July and became generally available on 31 August, and OCR 4.0 was retired on 30 September. It turns PDFs, Word, PowerPoint and OpenDocument files into text, and returns paragraph-level bounding boxes, labels for each block (titles, tables, equations, signatures and more) and confidence scores at page, block and word level.
It reads 170 languages across 10 language groups. It is offered through Mistral's API, Amazon SageMaker and Microsoft Foundry, and as a single container for enterprise customers who want to run it on their own infrastructure. Document AI uses the same endpoint to return JSON that matches a schema you provide.
The API costs $4 per 1,000 pages, $2 through the Batch API, and $5 per 1,000 pages for Document AI annotations.
How much does Mistral OCR 4.1 cost?
| Plan | Price | What you get |
|---|---|---|
| OCR API | $4 per 1,000 pages | Bounding boxes, block labels, confidence scores |
| Batch API | $2 per 1,000 pages | |
| Document AI annotations | $5 per 1,000 pages | Schema-matched JSON output |
$4 per 1,000 pages; $2 per 1,000 pages on the Batch API; Document AI annotations $5 per 1,000 pages. Self-hosted container for enterprise customers by arrangement.
Is Mistral OCR 4.1 worth it?
OCR 4.1 is the most complete document-reading API Mistral has shipped. The additions over OCR 3 matter for real pipelines: bounding boxes let you highlight where an answer came from, block labels separate tables from body text, and confidence scores let you send uncertain pages to a person instead of trusting them blindly.
Mistral's benchmarks are good but hedged. It reports 85.20 on OlmOCRBench, the top score among the systems it tested, 93.07 on OmniDocBench, and a 72% average win rate in a blind comparison over 600 documents in 12 languages. Mistral itself says competitor scores are its own reproductions and calls the results directional, and the per-competitor numbers appear only in charts. Users on Hacker News found it strong on typed scans and weaker than Claude on handwriting.
Price is the catch. At $4 per 1,000 pages it costs four times the $1 rate many guides still quote for older Mistral OCR, more than double some estimates for AWS Textract and Azure Document Intelligence, and well above Cohere's $1.50. The Batch API halves it, and for most archive work batch is the right choice.
For European companies that want accurate, layout-aware OCR with EU hosting or an on-premises container, it is a strong option. For cost-sensitive, high-volume jobs, open models such as PaddleOCR-VL or DeepSeek-OCR 2 on your own GPUs are far cheaper.
Where it earns its place
A legal or patent firm digitising case files
Extract text with page locations for citations
Use bounding boxes to link each extracted passage to its place on the page, and route low-confidence blocks to review. A 200,000-page archive costs about $400 on the Batch API.
A bank processing statements and forms
Turn documents into structured JSON
Document AI returns JSON matching your schema at $5 per 1,000 pages. Where data cannot leave the bank, ask Mistral for the self-hosted container.
A company building retrieval over multilingual documents
Index documents in many languages for search
OCR 4.1 covers 170 languages and feeds Mistral's open-source Search Toolkit. Block labels let you index tables and body text separately.
Before you commit
- Setup
- Call mistral-ocr-latest (now OCR 4.1) on the /v1/ocr endpoint, or /v1/batch for half price; also on Amazon SageMaker and Microsoft Foundry.
- Integrations
- Mistral API, Document AI structured output, Search Toolkit, Amazon SageMaker and Microsoft Foundry; Snowflake listed as coming.
- Data and privacy
- API terms follow Mistral's: free-tier data may be used for training unless you opt out, and pay-as-you-go users can opt out. A single-container self-hosted deployment keeps documents on your infrastructure for enterprise customers.
- Cost at scale
- A million pages cost $4,000 on the standard API or $2,000 on batch; Document AI annotations $5,000. Batch files up to 512 MB and 100,000 requests.
- Limits
- Documents only: not for audio, video or real-time work, and Mistral says not for medical, legal or high-stakes financial decisions. Scoring artefacts affect maths-heavy and multi-column pages.
What's good
- Paragraph bounding boxes, block labels and page, block and word confidence
- 85.20 on OlmOCRBench and 93.07 on OmniDocBench, per Mistral
- 170 languages; strong on low-resource languages in Mistral's tests
- Single-container self-hosting for enterprise customers
What's not
- $4 per 1,000 pages, four times older Mistral OCR pricing
- Competitor scores are Mistral's own reproductions
- Weaker on handwriting than Claude, per early users
- Self-hosting only through an enterprise agreement
Best for
Teams building document pipelines that need layout, confidence scores and EU or on-premises hosting more than the lowest cost per page.
Mistral OCR 4.1 alternatives
Mistral OCR 4.1: common questions
- Is Mistral OCR 4.1 worth it?
- Strong, layout-aware OCR with useful confidence scores and a self-hosting option, at four times the price of its predecessor and of some rivals. We rate it 4.2 out of 5.
- How much does Mistral OCR 4.1 cost?
- $4 per 1,000 pages; $2 per 1,000 pages on the Batch API; Document AI annotations $5 per 1,000 pages. Self-hosted container for enterprise customers by arrangement.
- What are the downsides of Mistral OCR 4.1?
- $4 per 1,000 pages, four times older Mistral OCR pricing; competitor scores are Mistral's own reproductions; weaker on handwriting than Claude, per early users; self-hosting only through an enterprise agreement.
- Who is Mistral OCR 4.1 best for?
- Teams building document pipelines that need layout, confidence scores and EU or on-premises hosting more than the lowest cost per page.
- How much does Mistral OCR 4.1 cost for a team or in production?
- A million pages cost $4,000 on the standard API or $2,000 on batch; Document AI annotations $5,000. Batch files up to 512 MB and 100,000 requests.
- Is Mistral OCR 4.1 safe for company data?
- API terms follow Mistral's: free-tier data may be used for training unless you opt out, and pay-as-you-go users can opt out. A single-container self-hosted deployment keeps documents on your infrastructure for enterprise customers.
- What are the best Mistral OCR 4.1 alternatives?
- The closest alternatives we have reviewed are PaddleOCR-VL-1.6, DeepSeek-OCR 2, Unlimited-OCR and Gemini 3.5 Flash-Lite.
Answers checked Oct 25, 2026.
Make Mistral OCR 4.1? Show this rating on your site
