Mistral AI · 2026-06-23 · major
Mistral OCR 4 — 170-language document model with bounding boxes and confidence scores
Mistral OCR 4 extracts text plus per-block bounding boxes, type labels, and confidence scores across 170 languages, scoring 85.20 on OlmOCRBench and 93.07 on OmniDocBench at $4 per 1,000 pages.
Mistral's new document model returns structured pages with boxes, block types, and per-word confidence at $4 per 1,000.
Key specs
| Price | $4 / 1K pages |
|---|---|
| Olm ocrbench | 85.20 |
| Omni doc bench | 93.07 |
| Languages | 170 |
Quick facts
| Maker | Mistral AI |
|---|---|
| Model name | mistral-ocr-4-0+2 |
| Languages | 170 across 10 language groups |
| Output | Text, bounding boxes, block types, confidence scores |
| Price (API) | $4 / 1,000 pages |
| Price (Batch) | $2 / 1,000 pages |
| Availability | Mistral Studio, Amazon SageMaker, Microsoft Foundry, self-hosted |
Pricing
| API | $4 / 1,000 pages |
|---|---|
| Batch API · 50% discount on async jobs | $2 / 1,000 pages |
| Document AI · Adds structured annotations and Document QnA | $5 / 1,000 pages |
What is it?
Mistral OCR 4 turns a scanned or rendered page into a structured object instead of a flat text dump. Each detected block ships with a bounding box, a type label such as title, table, equation, or signature, and confidence scores at both the page and word level. The same model covers 170 languages across 10 language groups in one container.
How does it work?
A single Mistral document model processes the page in one pass, emitting paragraph-level boxes alongside the classified block tree. Confidence is exposed inline so calling agents can route low-score regions to a human or a second-pass model. The Document AI tier layers structured annotations and Document QnA on the same OCR output through the `/v1/ocr` endpoint.
Why does it matter?
Teams building RAG, claim processing, or contract workflows previously stitched OCR, layout detection, table parsing, and signature spotting from separate vendors. Mistral OCR 4 collapses that pipeline into one model and one bill at $4 per 1,000 pages, with self-hosting for regulated workloads. The OlmOCRBench score of 85.20 and OmniDocBench score of 93.07 put it at the top of public OCR leaderboards.
Who is it for?
document automation, RAG, and AI-agent teams
Frequently asked questions
- How is Mistral OCR 4 different from OCR 3?
- Mistral OCR 4 returns a structured representation of each page instead of just clean text and tables. Every block carries bounding-box coordinates, a type label such as title, table, equation, or signature, and inline per-page and per-word confidence scores so downstream pipelines can route low-confidence regions for review.
- How much does Mistral OCR 4 cost?
- Mistral OCR 4 costs $4 per 1,000 pages through the standard API and $2 per 1,000 pages through the Batch API, a 50% discount for async jobs. The Document AI tier, which adds structured annotations and Document QnA on top of OCR, is $5 per 1,000 pages.
- How many languages does Mistral OCR 4 support?
- Mistral OCR 4 supports 170 languages across 10 language groups, with a 0.98 score on Mistral's internal Crawl Multilingual evaluation. The single-container deployment ships every language in the same model, so multilingual documents do not need a language switch.
- Where can you run Mistral OCR 4?
- Mistral OCR 4 is available via the Mistral Studio API and through Amazon SageMaker and Microsoft Foundry, with Snowflake Parse Document support listed as coming. Enterprise customers can also self-host the model in a single container, which keeps document data inside their own environment.
Try it
curl https://api.mistral.ai/v1/ocr -H "Authorization: Bearer $MISTRAL_API_KEY" -d '{"model":"mistral-ocr-latest"}'