Mistral OCR 4

Mistral OCR 4 Online for Layout-Aware Document OCR

Mistral OCR 4 is the stronger choice when the page layout matters, not just the words. It adds bounding boxes, block classification, and confidence scores for RAG citations, redaction, review queues, form extraction, and audit-ready document pipelines.

Billing note: recognition is charged by processed page, not by uploaded file. OCR 4 uses 2 credits per processed page.

Try OCR 4 Compare OCR 3

OCR 4 advantage

Layout-aware structured OCR

Site credit usage

2 credits / page

Best when

You need layout, labels, and confidence

Strength

RAG, citations, redaction, review

Upload Your Document

Supports PDF, JPG, PNG and more

Checking login status...

What is new in OCR 4?

Bounding boxes and block labels

OCR 4 is the better choice when automation needs to know where titles, tables, images, captions, headers, footers, and signatures appear on the page.

Confidence scores for review

Use page-level or word-level confidence signals to flag risky text for quality control, sampling, and human-in-the-loop review queues.

Structured table extraction

OCR 4 is strongest when tables need to stay machine-readable and connected to surrounding layout, captions, and document context.

Multilingual document workflows

Use OCR 4 for global archives, mixed-language PDFs, and document pipelines where structured output matters across language groups.

Mistral OCR 3 vs OCR 4: strengths and credit usage

OCR 4 uses more credits per page because it returns richer structured output such as bounding boxes, block labels, and confidence scores. OCR 3 remains the better choice when lower credit usage matters more than layout-aware automation.

Decision point	OCR 3	OCR 4
Site credit usage	1 credit / page	2 credits / page
Primary advantage	Lower-cost markdown, text, and table OCR	Layout-aware OCR with richer structured metadata
Best fit	Bulk conversion and cost-sensitive extraction	RAG, citations, redaction, review, forms, and audit workflows
Advanced metadata	Use when coordinates and confidence are optional	Best fit for bounding boxes, block labels, and confidence scores

Best-fit workflows

Semantic chunking for RAG and enterprise search

Source-grounded citations with page and block references

Invoice, receipt, form, and contract extraction

Redaction workflows that need precise source regions

Compliance review with confidence-based sampling

Large archive digitization through batch OCR

When should you choose OCR 4?

Choose OCR 4 when you need layout-aware extraction, source references, reviewable confidence data, or document pipelines that depend on where content appears on the page. Choose OCR 3 when lower cost matters more than the newest structured extraction features.

This website now defaults to OCR 4 on the homepage processor, while still letting you switch back to OCR 3 before running recognition.

Try Mistral OCR 4 online

Upload a PDF or image, keep OCR 4 selected, and download the recognized Markdown result. For advanced structured output such as blocks and confidence scores, OCR 4 gives the strongest foundation for future workflows.

Start OCR 4 recognition