Name: Mistral OCR Latest API
Brand: Mistral AI

Overview

Mistral OCR Description

Mistral OCR, developed by Mistral AI, is an advanced Optical Character Recognition (OCR) API designed for superior document understanding. It processes PDFs, images, and scanned documents, extracting text, tables, equations, and images with high accuracy while preserving document structure.

Technical Specifications

Performance Benchmarks

Mistral OCR leverages a transformer-based architecture with specialized attention mechanisms to understand document context and layout. It supports multimodal inputs (PDFs, images) and outputs structured formats like Markdown and JSON, optimized for integration with Retrieval-Augmented Generation (RAG) systems.

Context Window: Processes up to 1000 pages per request.
Benchmarks:
- Overall Accuracy: 94.89% (outperforms Google Document AI, Azure OCR, GPT-4o)
- Mathematical Expressions: 94.29%
- Multilingual Text: 89.55%
- Scanned Documents: 98.96%
- Table Recognition: 96.12%
Performance: Processes up to 2000 pages per minute on a single node.
API Pricing:
- $0.0013 per page
Limitations:
- Maximum file size: 50 MB
- Maximum page count: 1000 pages

Performance Metrics

Key Capabilities

Mistral OCR redefines document processing by combining AI-driven text extraction with deep layout understanding, supporting thousands of languages and complex document elements like LaTeX, tables, and images. It outputs structured data for seamless integration into AI workflows.

High-Accuracy Text Extraction

Achieves 94.89% overall accuracy, outperforming competitors in extracting text from scanned documents, handwritten notes, and multilingual content, ensuring reliable data for downstream applications.

Multimodal Document Understanding

Processes PDFs and images, recognizing interleaved images, tables, charts, and mathematical equations, preserving their context and relationships in structured Markdown or JSON outputs.

Multilingual Proficiency

Supports thousands of languages with 99.02% fuzzy match accuracy, making it ideal for global organizations processing diverse document sets, from Hindi to Chinese.

Structured Output and Layout Preservation

Retains document hierarchy (headers, paragraphs, lists, tables) in outputs, enabling AI-ready formats for RAG systems, search indexing, and automation workflows.

Doc-as-Prompt Functionality

Allows users to query specific document content or extract structured data using AI-driven prompts, enhancing precision in information retrieval and analysis.

High-Speed Processing

Handles up to 2000 pages per minute, optimized for large-scale document repositories, reducing processing time for enterprises and research institutions.

Self-Hosting for Data Privacy

Offers on-premises deployment for organizations with strict security needs, ensuring sensitive data remains within private infrastructure.

Optimal Use Cases

Research and Academia: Digitizing scientific papers with equations and charts into AI-ready formats.
Business and Finance: Processing invoices, contracts, and reports for structured data extraction.
Legal and Compliance: Converting filings and records into searchable, indexed formats.
Education: Transforming lecture notes and textbooks into accessible digital content.
Customer Service: Indexing manuals to reduce response times and improve satisfaction.

Comparison with Other Models

Mistral OCR excels in document understanding, surpassing traditional and AI-based OCR solutions:

vs. Gemini 2.5 Flash: Superior in OCR accuracy (94.89% vs. ~88.49%) and table recognition, but lacks Gemini’s general multimodal reasoning.
vs. Google Document AI: Higher accuracy in math (94.29% vs. ~90%) and multilingual text (89.55% vs. ~85%), with faster processing (2000 vs. ~1000 pages/min).
vs. Azure OCR: Better layout preservation and structured outputs, though Azure offers broader enterprise integrations.
vs. GPT-4o: Outperforms in scanned documents (98.96% vs. ~95%) and equations, but GPT-4o is more versatile for non-OCR tasks.

Code Samples

Limitations

Hallucinations: May guess missing or unclear text, risking errors in critical applications (e.g., legal, financial).
No Document Classification: Requires additional systems for organizing extracted data.
Text Misclassification: Some pages may be treated as images, leading to incomplete extraction.
File Constraints: Limited to 50 MB files and 1000 pages per request.

API Integration

Accessible via AI/ML API. Supports Python, JavaScript, and cURL, with structured outputs in JSON/Markdown. Documentation available here.

‍

Uptime and performance

Uptime (24h): 100%
Uptime (7d): 100%
Availability (24h): 100%
Error rate (24h): 0%
End-to-end latency (24h median): 3.4s

Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.

Use cases

Enterprise AI workflows for document processing, compliance, customer service automation
Research institutions digitizing and structuring scientific papers and journals
Cultural and historical heritage digitization and preservation
Legal, educational, and technical document automation and analysis
AI-enabled content indexing for search engines, chatbots, and knowledge management systems

How to use Mistral OCR Latest API

Install any OpenAI-compatible SDK, point it at api.aimlapi.com/v1, and set the model to mistral/mistral-ocr-latest.

import requests

headers = {"Authorization": "Bearer " + AIMLAPI_KEY}
res = requests.post(
    "https://api.aimlapi.com/v1/ocr",
    headers=headers,
    json={
      "model": "mistral/mistral-ocr-latest",
      "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-latest",
    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-latest",
    "document": "https://example.com/invoice.pdf"
  }'

OpenAI-compatible — swap the base URL and it works with your existing SDK.