# Mistral AI Launches New Document Recognition Model Mistral OCR 4

- Category: Flash (快讯)
- Published: 2026-06-26T11:56:00.000Z
- Source date: 2026-06-26
- HTML: https://www.88z.ai/en/news/flash-20260626-26e22a06dddff859
- Markdown: https://www.88z.ai/en/news/flash-20260626-26e22a06dddff859.md
- Other language: https://www.88z.ai/news/flash-20260626-26e22a06dddff859.md

Mistral AI has released Mistral OCR 4, a new document recognition model supporting 170 languages. It scored 93.07 on the OmniDocBench benchmark, outperforming some competitors in human preference. The model provides bounding boxes, region classification, and confidence scores, making it suitable for downstream tasks like RAG.

## At a glance

- Mistral OCR 4 supports 170 languages and scored 93.07 on OmniDocBench
- The model provides bounding boxes and confidence scores to optimize downstream tasks like RAG

## The update

Mistral AI has officially launched its latest document recognition model, Mistral OCR 4. This model features robust multilingual capabilities, handling 170 languages across 10 language families. It achieved an impressive score of 93.07 on the authoritative OmniDocBench benchmark. Compared to market competitors such as GPT 5.5 Pro and Gemini 3.1 Pro Preview, its generated outputs are preferred by human evaluators. In addition to high-quality text extraction, the model simultaneously provides structured information including bounding box coordinates, region classification, and confidence scores. These features enable better support for downstream workloads like retrieval-augmented generation (RAG) semantic chunking, offering developers more refined document processing capabilities.

## Sources

- [IT之家](https://www.ithome.com/0/968/835.htm) (2026-06-26)

This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.