Match each agent to the best-fit AI engine while keeping data governed.
Effective May 28, 2026, Microsoft added Mistral Medium 3.5 as a new model provider in Copilot Studio.
Summary
- Microsoft added Mistral Medium 3.5 as a selectable model provider in Microsoft Copilot Studio.
- The model targets agent building and orchestration, and is available worldwide in early release environments.
- European Union (EU) organizations can keep data processing in-region and cut procurement overhead.
- Access needs a two-step admin opt-in across the Microsoft 365 admin center and the Power Platform admin center.
- Microsoft marks the model experimental and advises non-production use during testing.
- Immediate action: review the data terms that apply to external, third-party models before enabling them.
What is changing?
Microsoft expanded the model lineup in Copilot Studio. Mistral Medium 3.5 now appears in the model picker next to existing options, so makers can choose the model that fits each agent. The model is available worldwide for customers in early release environments, and Microsoft classifies it as experimental for now.
The new option is off by default. It becomes visible to makers only after an administrator opts in at the tenant level and enables external model providers for the selected environments. Existing agents and workflows stay unchanged unless an admin turns the option on.
What is Microsoft Copilot Studio?
Microsoft Copilot Studio is Microsoft’s low-code environment for building, orchestrating, and governing AI agents. It sits on top of Microsoft Power Platform and connects to enterprise data and systems. Through a multi-model selector, teams can pick different models for different agents while keeping orchestration, governance, and lifecycle management in one place.
Customer benefits at a glance
- Better answers per use case, because teams match the model to the task.
- Easier compliance conversations with EU stakeholders on where data is handled.
- Lower switching cost when trying a new model.
- Consistent oversight, since nothing leaves your admin perimeter.
- Confident experimentation ahead of production use.
Key capabilities
- Built for long-horizon tasks and reliable multi-tool calling.
- Produces structured output for downstream code.
- Reasoning effort is configurable per request, from quick replies to complex agent runs.
- Selectable directly in the Copilot Studio model picker once enabled.
Why governed model choice matters now
More organizations are moving agents from pilots into production. A single default model rarely fits every language, region, or workload. Governed model choice reduces risk, because third-party models stay inside known admin controls. It increases clarity for EU teams that must show where data is processed. And it saves time, because evaluation happens without separate vendor onboarding.
What organizations should do now
- Review which agents could benefit from an alternative model, and list them.
- Validate the data-processing terms for external models against your compliance policy.
- Enable the two required switches for a single test environment first.
- Monitor experimental-model usage and keep it out of production until testing completes.
- Update internal governance guidance so makers know when external models are allowed.
Ready to review your Copilot Studio licensing?
SCHNEIDER IT MANAGEMENT helps you assess the licensing and governance impact of new model options in Copilot Studio. Our experts clarify what external models mean for your agreements, data terms, and compliance posture. Contact us to align your Microsoft licensing with your AI agent strategy.
Sources
For the announcement of Microsoft Copilot Studio: More Model Choice with Mistral Medium 3.5, please visit: https://www.microsoft.com/en-us/microsoft-copilot/blog/copilot-studio/mistral-joins-copilot-studios-growing-lineup-of-model-providers/.
For useful software licensing information on Microsoft products, please visit: https://www.schneider.im/software/microsoft/.