

Mistral OCR 4
#10 in Transkription (STT)Mistral AI · v4 · siet 2026-06-23 · 3× · tolest 29. Juni 2026
Mistral OCR 4 is a compact, document-focused OCR model released by Mistral AI on June 23, 2026. Beyond plain text extraction, it returns structured output with paragraph-level bounding boxes, typed block classification (titles, tables, equations, signatures), and per-page and per-word confidence scores. The model supports 170 languages across 10 language groups, accepts PDF, DOC, PPT, and OpenDocument formats, and can be deployed as a single container for fully self-hosted use. On the public OlmOCRBench it achieves the top score of 85.20; independent annotators preferred it in 72% of blind comparative evaluations against leading competitors.
Features
| Price Tier | API: $4 / 1,000 pages; Batch API: $2 / 1,000 pages (50% discount); Document AI (Mistral Studio): $5 / 1,000 pages; Self-Hosted Enterprise: price upon request |
| Processing Speed (x Realtime) | Up to 2,000 pages per minute on a single GPU |