Mistral OCR 3 is an advanced optical character recognition and document understanding model designed to process PDFs, scanned images, forms, and other document formats. Its primary goal is not only to recognize text but also to preserve the context and structure that give documents meaning.
This distinction is important. Traditional OCR systems often return large blocks of plain text, forcing developers to spend significant time reconstructing layouts and extracting relevant information. Mistral OCR 3 approaches the problem differently by understanding the relationships between headings, paragraphs, tables, forms, and annotations.
OCR Pages:
Annotated Pages:
Documents rarely arrive in perfect condition. Many contain handwritten notes, poor-quality scans, rotated pages, stamps, signatures, or complex formatting that can confuse conventional OCR tools. Mistral OCR 3 has been designed specifically for these real-world scenarios. The model performs reliably across a wide range of document types and can accurately process content that would traditionally require manual review.
Its ability to understand visual structure makes it particularly effective for business documents where context matters just as much as the text itself. Whether a page contains financial tables, legal clauses, application forms, or technical documentation, the model can preserve the organization of information while extracting the content.
One of the most significant improvements in Mistral OCR 3 is its handwriting recognition capability. Handwritten documents have historically been one of the most difficult challenges in document digitization. Variations in writing style, spacing, and legibility often lead to inaccurate results. Mistral OCR 3 addresses this issue by leveraging modern multimodal AI techniques that allow it to interpret handwritten content with a much higher degree of accuracy.
This makes the model suitable for processing annotated documents, handwritten forms, archived records, meeting notes, and mixed-content files where printed and handwritten information appear together on the same page.
Tables are often where traditional OCR solutions struggle the most. Financial statements, invoices, reports, and scientific publications frequently contain complex table structures that lose meaning when converted into plain text. Mistral OCR 3 is designed to preserve these structures during extraction. Instead of flattening rows and columns into a sequence of words, the model reconstructs relationships between cells, headers, and nested sections.
The result is a much cleaner representation of data that can be directly integrated into analytics systems, databases, spreadsheets, or AI applications. This level of structural understanding helps organizations reduce processing errors while improving the quality of downstream workflows.
Global organizations require document processing systems that work consistently across different languages and formats. Mistral OCR 3 has been optimized for multilingual environments, enabling teams to process international documents without relying on separate OCR solutions for each region.
Whether handling contracts, customer records, research materials, or regulatory documents, the model is designed to maintain accuracy while preserving the original structure of the content.
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
Mistral OCR 3 is documented as suited for extracting text and images into markdown for downstream agents and knowledge systems, and for automated parsing of forms, invoices, and operational documents. It is also described as useful for end-to-end document understanding pipelines and for digitizing handwritten or historical documents. The model is additionally cited for document-to-knowledge transformation applications, including turning company archives and technical or scientific reports into clean, searchable text.
import requests headers = {"Authorization": "Bearer " + AIMLAPI_KEY} res = requests.post( "https://api.aimlapi.com/v1/ocr", headers=headers, json={ "model": "mistral/mistral-ocr-3", "document": "https://example.com/invoice.pdf" }, ).json() print(res)
const res = await fetch("https://api.aimlapi.com/v1/ocr", { method: "POST", headers: { Authorization: `Bearer ${AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ model: "mistral/mistral-ocr-3", document: "https://example.com/invoice.pdf", }), }).then((r) => r.json()); console.log(res);
curl -s https://api.aimlapi.com/v1/ocr \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "mistral/mistral-ocr-3", "document": "https://example.com/invoice.pdf" }'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Output | |
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Mistral OCR 3 This page | Document text extraction | |||
| Document text extraction | ||||
| Document text extraction | ||||
| Document text extraction |
Mistral OCR 3 takes document, image as input and returns text.
Mistral OCR 3 is priced at $2.6 / 1K pages.
Mistral OCR 3 is billed per generation — a fixed charge per output rather than by prompt length.
Mistral OCR 3 was built by Mistral AI.
Use mistral/mistral-ocr-3 as the model id on AI/ML API.
Yes. Mistral OCR 3 is served through AI/ML API, so the same key and endpoint format used for other models applies.
The model ID is mistral/mistral-ocr-3.
It is designed for PDFs and images, extracting text and embedded images and producing markdown output with HTML-based table reconstruction.
It is fully backward compatible with Mistral OCR 2.
