OCR 4.1
AI model · Computer vision · Image → Text by Mistral AI
Premier document extraction and understanding model for structured data.
OCR 4.1 is Mistral AI's premier document extraction model that turns unstructured files into structured data. It features native paragraph-level bounding boxes and structural labels, making it ideal for processing complex tables, invoices, and handwriting at scale.
Model specs
- Model family
- OCR
- Version
- 4.1
- Released
- 2026-07-16
- Input
- Image, Files
- Output
- Text, Files
- Standard OCR
- $4 per 1000 pages
- Annotated OCR
- $5 per 1000 pages
- Capabilities
- Structured output, Batch
- Launched
- 2026-07-16
Category: Computer Vision
- Modalities
- Image, Document → Text, JSON
- Sizes and variants
- OCR 4.1 (v4.1), OCR 3 (v25.12)
- Access
- API, Mistral Studio, Batch API
- Released
- 2026-07-16
- Version
- 4.1
- Model ID
- mistral-ocr-4-1
- Native paragraph-level bounding box extraction
- Structural block labels and block-level confidence scores
- Optimized for tables, invoices, and handwriting
- Powers the Mistral Document AI stack
- $4 per 1,000 pages pricing
Overview
OCR 4.1 is Mistral AI's premier document understanding and extraction model, released in July 2026. It is designed for enterprises and developers who need to transform unstructured documents into structured, machine-readable data. It is optimized to handle complex layouts, including tables, invoices, and handwritten text. Delivered via the Mistral API, OCR 4.1 powers the company's Document AI stack and supports high-volume workloads through batch processing. The model is priced at $4 per 1,000 pages, providing a cost-effective solution for large-scale automated transcription and data extraction.
Document Understanding
- BBox Extraction: The model extracts text and structural elements from documents with native paragraph-level bounding box extraction.
- Structural Block Labels: It identifies and labels structural blocks within documents to preserve the original layout and hierarchy.
- Block-level Confidence: The model provides block-level confidence scores to indicate the reliability of the extracted data.
- Structured Data Output: It transforms unstructured documents like contracts, tables, and invoices into structured JSON data.
Extraction Capabilities
- Complex Layouts: The model is designed to handle complex document layouts including multi-column text and embedded tables.
- Table Processing: It supports the extraction and understanding of tabular data within documents.
- Handwriting Support: The model is capable of processing and transcribing handwritten text within documents.
- Multilingual Support: It provides multilingual support for document processing across various languages.
Deployment and Integration
- OCR API: The model is available through the /v1/ocr endpoint for integration into applications.
- Batch Processing: Users can process high-volume document workloads using the /v1/batch endpoint.
- Document AI Stack: The model serves as the foundation for the Mistral Document AI stack for enterprise-grade processing.
Use cases
- Financial institutions use OCR 4.1 to automate the extraction of data from complex loan agreements and invoices.
- Legal teams process large volumes of contracts to identify and label structural blocks for faster document review.
- Logistics companies extract data from shipping manifests and bills of lading to streamline supply chain operations.
- Public sector agencies use the model to transcribe historical documents and handwritten forms into structured digital records.
Specifications
- Modalities
- Image, Document → Text, JSON
- Tasks
- OCR, document understanding, structural extraction, layout analysis
- Sizes and variants
- OCR 4.1 (v4.1), OCR 3 (v25.12)
- Access
- API, Mistral Studio, Batch API
- Released
- 2026-07-16
- Version
- 4.1
- Model ID
- mistral-ocr-4-1
- Status
- General Availability (GA)
- Tier
- Premier
- Pricing Unit
- Per 1,000 pages
- Output Format
- JSON, Markdown
- Features
- BBox Extraction, Structural Labels, Confidence Scores
How it works
- Access the API: Access the model via the Mistral API or through the Mistral Studio interface.
- Submit Documents: Submit documents such as PDFs or images to the /v1/ocr endpoint for processing.
- Extraction and Analysis: The model analyzes the document to extract text, identify structural blocks, and generate bounding boxes.
- Receive Structured Data: Receive the processed data in a structured JSON format ready for use in downstream applications.
Pricing
- Standard Input
- $4 per 1,000 pages
- Annotated Input
- $5 per 1,000 pages
FAQ
How is OCR 4.1 priced?
Mistral OCR 4.1 is priced per 1,000 pages, with a standard rate of $4 per 1,000 pages and an annotated rate of $5 per 1,000 pages.
Can I process documents in batches?
Yes, OCR 4.1 supports batch processing through the /v1/batch endpoint, allowing for high-volume document extraction.
What is the difference between OCR 4.1 and OCR 3?
OCR 4.1 is Mistral's latest premier model, while OCR 3 remains available for existing integrations and production workloads.
Added by AntSpace from public sources (docs.mistral.ai).