AzureFeeds

In production AI systems, different problems require different model capabilities. Document ingestion, reasoning, coding, and automation workflows rarely rely on a single model. They depend on selecting the right tool for each layer of the stack. 

Microsoft Foundry continues to onboard Mistral models to address these complementary needs. 

This week introduces two additions:  

• Mistral Document AI (with OCR 4), arriving today: brings structured document understanding into production pipelines 

• Mistral Medium 3.5, arriving tomorrow: adds an open-weight general-purpose model for reasoning, coding, and agentic workloads 

Together, these releases broaden the range of capabilities available in Foundry, giving developers more flexibility to build, combine, and deploy models across different AI workloads, from document processing to customizable, large-scale AI systems. 

Build more reliable document pipelines with OCR 4 

In production systems, OCR is rarely the end goal. It’s the first step in a larger pipeline. 

Developers building RAG systems, agent workflows, or document automation often spend more time reconstructing layout and structure than on the downstream AI logic itself. Flattened text outputs make it difficult to: 

• Preserve relationships between elements (tables, headers, sections) 

• Trace outputs back to source locations for validation or citations 

• Build reliable ingestion pipelines for large-scale document sets 

Mistral Document AI with OCR 4 directly targets this gap. 

It introduces structure-aware document processing, where outputs are not just text but structured representations of the document, including layout, classification, and spatial context. 

“Documents remain one of the largest untapped sources of enterprise intelligence. With Mistral Document AI and OCR 4 available in Microsoft Foundry, organizations can move beyond text extraction and unlock structured, actionable data that powers automation, compliance, and AI-driven decision making at scale.” 
— Marjorie Janiewicz, Chief Revenue Officer, Mistral AI 

What changes with Mistral Document AI (with OCR 4)

OCR 4 enables a different approach to document pipelines: 

• True structured document understanding 

Introduces paragraph-level bounding boxes and block classification, preserving layout and enabling downstream use cases like citations, validation, and form extraction. 

• Quality and reliability improvements 

According to Mistral, OCR 4 delivers higher quality outputs with better handling of complex formats and formatting edge cases, compared to its predecessor. 

• Multilingual at global scale 

Expands to 170 languages, supporting mixed-language documents and global workflows 

• Supports enterprise workflows and compliance 

Enables structured outputs, traceability, redaction workflows, and human-in-the-loop validation  

• Efficient at scale 

Optimized for high-throughput, batch document processing, and deployable in a single container, OCR 4 enables cost efficiency and scalability in enterprise environments. 

Where developers are already applying it 

OCR 4 serves as a foundational layer for Document AI, enabling developers to build a range of production-ready workflows. It supports RAG pipelines by generating structured, citation-ready document chunks, powers agentic workflows where systems act on document elements (such as extracting fields or validating signatures), and enables large-scale enterprise ingestion pipelines for search, knowledge bases, and analytics. In regulated scenarios, it also strengthens compliance and audit workflows by providing structured, traceable outputs that support validation and human-in-the-loop review. When deployed in Microsoft Foundry, these capabilities integrate directly into broader AI systems, allowing document understanding to flow into downstream applications without requiring extensive pipeline rework. 

Read Mistral’s blog for more information. 

Mistral Document AI with OCR 4 Pricing and deployment 

Mistral Medium 3.5: open-weight reasoning and agentic AI 

Mistral Medium 3.5 expands the Foundry model catalog with a general-purpose model optimized for reasoning, coding, and agent workflows. 

Medium 3.5 is a 128B parameter dense model with a 256K token context window, supporting both text and image inputs. It is also Mistral’s first fully merged flagship model, combining instruction-following, reasoning, and coding capabilities into a single system, reducing the need to orchestrate multiple specialized models.  

What differentiates Medium 3.5 is its open-weight design. Available under a modified MIT license, it gives organizations more cost-efficiency and ability to customize to suit specific needs. The customization is especially compelling for governments, financial services, healthcare, and global enterprises that require strict data residency, compliance, and auditability without giving up advanced AI capabilities.  

Read Mistral Medium 3.5 blog to learn more. 

Mistral Medium 3.5 Pricing and deployment 

Why use Document AI (with OCR 4) and Medium 3.5 on Microsoft Foundry 

These models can be powerful when used together in a single AI architecture. 

Developers can: 

• Use OCR 4 to convert documents into structured, pipeline-ready data 

• Use Medium 3.5 to support reason over that data, powering copilots, agents, and automation 

All within Microsoft Foundry’s unified platform. 

“The launch of Mistral Medium 3.5 and OCR 4 in Microsoft Foundry reflects our shared commitment with Microsoft to deliver powerful, enterprise-ready AI. Customers can now access leading AI capabilities through a trusted platform they already use today.”

— Dion Ubert, Head of Microsoft Partnership at Mistral AI

Foundry provides the foundation to run these workloads at enterprise scale and within regulated environments. Teams can process large document volumes with OCR 4 while leveraging models like Mistral Medium 3.5 for downstream reasoning and automation, both within a platform that enforces governance, security, and deployment control. With support for flexible deployment methods, organizations can meet strict data residency, auditability, and compliance requirements while scaling AI systems confidently across production workloads. 

Get started today 

Getting started with Mistral on Microsoft Foundry is straightforward: 

1. Create or access your Microsoft Foundry workspace 


2. Browse the model catalog and select: mistral-ocr-4-0 for document workflows; Mistral Medium 3.5 for reasoning and agentic workloads


3. Deploy the model using real-time endpoints or serverless API. Check out Foundry documentation here.


4. Integrate into your applications: from document pipelines to copilots and automation systems 

With OCR 4 unlocking structured document intelligence and Medium 3.5 enabling efficient, flexible deployment, organizations can start building end-to-end, product-ready AI solutions.