AI LaunchJuly 18, 2026 · Mistral AI
Mistral OCR 4 returns bounding boxes, not just text
Mistral OCR 4 ships bounding boxes, block types, and per-word confidence scores at $2 per 1,000 pages in batch mode.
What launched
Mistral released OCR 4 on June 23, 2026. The model returns bounding boxes, block classification (titles, tables, equations, signatures, and more), and inline confidence scores alongside extracted text. It supports 170 languages across 10 language groups, accepts PDF, DOC, PPT, and OpenDocument, and runs in a single container for self-hosted deployment. Independent annotators preferred OCR 4 over every competing system tested on 600+ documents across 12+ languages, with win rates averaging 72%. It scored 85.20 on OlmOCRBench. API pricing is $4 per 1,000 pages, with a 50% Batch-API discount bringing it to $2. Document AI in Mistral Studio, the no-code path to the same engine, is $5 per 1,000 pages. OCR 4 is also positioned as an ingestion component for the Mistral Search Toolkit, feeding citation-ready inputs into RAG pipelines. Aidan Donohue, AI Engineer at Rogo, said: "We benchmarked Mistral OCR 4 against the leading agentic document parsers across a chart and figure dense financial QA dataset and reached equivalent accuracy at roughly 8x lower cost and 17x lower latency."
For designers
Document viewers can now highlight the exact region a citation came from, because the OCR layer returns coordinates and a block type for every element. That shifts the design work from styling extracted text to designing verification UI — the click-through from an answer to the signature, table cell, or clause it came from, with a confidence score attached.