Mistral OCR 4

Mistral AI

Mistral's document-intelligence model, released June 2026. It parses PDF, DOC, PPT and OpenDocument files across 170 languages in 10 language groups, handling interleaved imagery, mathematical expressions, tables and complex layouts. Every request returns markdown-structured text alongside bounding boxes, block types and confidence scores.

Strengths

93.07 on OmniDocBench and 85.20 on OlmOCRBench
170 languages across 10 language groups, including low-resource scripts
Handles tables, mathematical expressions, LaTeX and interleaved imagery
Returns bounding boxes, block types and confidence scores, not just flat text
Human annotators preferred it in a 72% average win rate against systems tested
Reported equivalent accuracy at roughly 8x lower cost and 17x lower latency than agentic document parsers on financial QA

Caveats

Proprietary - Mistral API only, though on-premises deployment is offered
A document-parsing model, not a general chat model
Priced per page rather than per token, so cost scales with document volume

Vendor: Mistral AI
License: Proprietary
Release Date: 2026-06-23
Modalities: text
image

Benchmark	Score	Max
OmniDocBench	93.1	100.0
OlmOCRBench	85.2	100.0

Capabilities

Vision

Audio

Video

Tool Use

Resources

API Documentation

Pricing

$4 per 1,000 pages via the API, $2 per 1,000 pages on the Batch API (50% discount), and $5 per 1,000 pages through Document AI.