Microsoft adds Mistral Large 3 model to Azure Foundry

Microsoft is putting another heavyweight model into its cloud arsenal. The company says Mistral Large 3, an Apache-licensed open-weight system, is now available in Azure through Microsoft Foundry for full production use.

The model targets enterprises that want transparent, flexible AI they can deploy across a mix of architectures without giving up reliability, multimodal reasoning or long-context performance.

Microsoft describes Mistral Large 3 as part of the “leading tier of globally available open models,” tuned for instruction following, long-context comprehension and multimodal reasoning, with an emphasis on stable behaviour in multi-turn conversations and complex inputs.

According to Microsoft, the model supports production assistants, retrieval-augmented applications, agentic systems and multimodal workflows. It can run inside Foundry, be exported for hybrid or on-premises setups depending on Mistral’s licensing terms, and operate in virtual private cloud, edge or sovereign environments.

Foundry serves as a unified entry point for Mistral Large 3 and Mistral Document AI, alongside other open and commercial models. It provides end-to-end evaluation, routing and observability across cost, latency, throughput and quality.

Built-in Responsible AI safeguards, content filters and auditability are included, as is tool-calling for building customer service bots, research agents, automation flows and enterprise copilots.

Microsoft says the model is suitable for use cases such as enterprise knowledge assistants, document intelligence, developer agents and multimodal customer experiences.

Mistral Large 3 is launching as a Global Standard deployment in the West US 3 Azure region at a price of $0.5 per 1 million input tokens and $1.5 per 1 million output tokens, and is available in public preview from December 2 2025. Microsoft says the addition expands Foundry’s “widest selection of open and frontier models” under a single enterprise-ready ecosystem.

The recap

• Microsoft Foundry adds Mistral Large 3 open-weight model on Azure.
• Model supports long-context, multimodal reasoning and agentic workflows.
• Priced at $0.5 input, $1.5 output per million tokens.