Mistral OCR 4.1: Technical Overview

Mistral OCR 4.1 serves as the foundational engine for the Document AI ecosystem. It provides advanced capabilities including native extraction of paragraph-level bounding boxes, the assignment of structural labels to blocks, and the provision of confidence scores for every extracted block.

🚀 Model Specifications

The current version, mistral-ocr-4-1, is available in Public Preview as part of the Premier tier.

• Lifecycle Date: July 16, 2026
• Status: Beta $\rightarrow$ Public Preview

💰 Pricing Structure

The cost of processing is calculated based on the volume of pages.

The pricing formula can be expressed as: $\text{Total Cost} = \left( \frac{\text{Total Pages}}{1000} \right) \times \text{Rate}$

---

🛠 Key Features & Endpoints

The following capabilities are integrated into the API:

• BBox Extraction: Located at /v1/ocr
• Standard OCR: Located at /v1/ocr
• Structured Annotations: Located at /v1/ocr
• Batch Processing: Located at /v1/batch

Workflow Visualization

Example API Implementation

To utilize the batching feature, you might use a request similar to this:

{
  "model": "mistral-ocr-4-1",
  "endpoint": "/v1/batch",
  "parameters": {
    "extract_bboxes": true,
    "include_confidence": true
  }
}

---

🖼 Visual Representation

🌐 Ecosystem Context

Alongside mistral-ocr-4-1, the following models are available within the environment:

• Z.ai GLM 5.2 (v 5.2)
• Shieldstral 1.0 (v 1.0)
• Mistral Medium 3.5 (v 26.04)