Mistral OCR 4 Released: Next-Gen Document Recognition Engine Supporting 170 Languages

French AI company Mistral AI released Mistral OCR 4, next-gen document recognition with bounding box detection, block classification (titles, tables, equations, signatures), and per-word confidence scores. Supports 170 languages across 10 language families including Chinese, Japanese, Arabic, Hebrew. Self-hostable via single Docker container for sensitive documents. Scored 85.20 on OlmOCRBench benchmark with 72% annotator preference rate. Priced at $4 per 1,000 pages (50% off for batch API). Ideal for enterprises processing contracts, invoices, forms at scale.