Microsoft has added Mistral Medium 3.5 to Copilot Studio for agent building and orchestration, giving organisations another model choice while keeping access behind two administrator approvals. The practical point is less “more models” than more control over where and how enterprise agents are run.
Microsoft AI Watch analysis
What happened
Microsoft says Mistral Medium 3.5 is available worldwide in early-release environments, with particular relevance for organisations in the European Union that want more flexibility around in-region data processing. The model is experimental, and Microsoft recommends using it for non-production work while testing and evaluation continue.
The model is designed for long-horizon tasks, multi-tool calls and structured output, according to Mistral’s description quoted by Microsoft. Its reasoning effort can be configured per request, allowing teams to choose between quicker replies and more involved agentic runs.
Access is not switched on by default. An administrator must first opt in through the Microsoft 365 admin centre, then enable external model providers in the Power Platform admin centre for selected environments. Once both switches are on, makers can choose “Mistral Medium 3.5 (Experimental)” in Copilot Studio’s model picker.
Why it matters
This gives enterprise teams a way to trial a non-Microsoft model inside Copilot Studio without immediately rebuilding their orchestration and governance arrangements elsewhere. For EU organisations, Microsoft says the addition can provide more flexibility over data processing and avoid some procurement overhead associated with adding a new provider.
The two-stage opt-in is the useful bit for administrators. It creates a deliberate path from pilot to wider use, rather than letting a new model quietly appear in every maker’s toolbox and hope the paperwork catches up. That is not glamorous, but neither is discovering an experimental model in production after the fact.
Our read
Microsoft is selling model choice with a governance wrapper, and in this case the wrapper is part of the story. Mistral Medium 3.5 may be useful for agents that need repeated tool calls or predictable structured output, but the supplied announcement does not provide independent performance comparisons or production evidence.
Teams should treat this as a controlled evaluation opportunity, not a ready-made reason to move important workloads. Start with non-production tasks, check regional processing requirements and test the model against the failure modes that matter to the organisation. Model variety is valuable only when the filing cabinet, audit trail and off switch keep pace.
What to watch
- Whether Microsoft moves Mistral Medium 3.5 beyond experimental status.
- Whether independent evaluations show an advantage on long-horizon and multi-tool tasks.
- Which regions, plans and Copilot Studio environments receive support next.
- Whether customers use the model choice to reduce cost, improve capability or meet data-location requirements.
Discussion spark: Should enterprise platforms make competing models available by default for experimentation, or is a two-stage administrator opt-in the minimum responsible gate for agent development?
Sources and evidence
- Mistral joins Copilot Studio’s growing lineup of model providers – Microsoft (16 September 2026, 16:26 UTC)
not affiliated with or endorsed by Microsoft