Mistral AI Debuts OCR 3 Model, Targeting Superior Document Intelligence

Details

• Mistral AI has launched OCR 3, its latest optical character recognition system designed for both structured and unstructured documents.
• Company testing shows it offers greater character-level accuracy and lower latency compared to established enterprise solutions and recent AI-based OCR offerings.
• OCR 3 is available now via a REST API and a no-code Document AI playground inside Mistral AI Studio.
• Pricing is published in a blog post, starting around $0.0003 per page, alongside throughput numbers for documents such as invoices, IDs, and multi-language scans.
• The product supports features like table extraction, document layout parsing, and automatic redaction—all accessible through the same API endpoint as Mistral’s language models.
• This marks Mistral’s first standalone perception tool, broadening its scope beyond large language models like Mixtral and positioning the company as a comprehensive document intelligence platform.

Impact

This move intensifies competition with leaders like Google Cloud Vision and AWS Textract by claiming better accuracy at a lower cost. If adopted widely, it could help drive uptake of sovereign, GDPR-compliant AI across Europe and prompt rivals to accelerate their integration of multimodal capabilities.