Mistral AI released OCR 4 today, a new optical character recognition model that marks a significant leap in document processing. According to the announcement, the system outperformed every leading OCR competitor in human evaluations—independent annotators preferred it seventy-two percent of the time. Beyond extracting text from images or PDFs, OCR 4 also identifies document structure: it returns bounding boxes pinpointing where text appears, classifies block types like tables and titles, and flags confidence scores for each extraction. The model supports one hundred seventy languages and runs self-hosted in a single container, appealing to organizations that need to keep document data on premises. Mistral positioned OCR 4 as a building block for enterprise search and retrieval-augmented-generation pipelines, integrated with the company's Search Toolkit.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton