Microsoft 365 Copilot is no longer a single-model product. Anthropic's Claude Opus 4.8 is now available directly inside M365 Copilot — meaning users can choose between OpenAI's GPT models and Anthropic's Claude without leaving Microsoft 365. For IT admins and decision-makers, this is a significant architectural shift worth understanding before your users start asking questions about it.
What Has Actually Changed
Until recently, Microsoft 365 Copilot ran exclusively on OpenAI models. That changed in April 2026 when Claude Opus 4.7 became available in Copilot Cowork (Frontier), and again in May 2026 when Claude Opus 4.8 started rolling out to Copilot Chat, Excel, PowerPoint, and Copilot Studio.
This is not a hidden backend switch Microsoft makes automatically. Users actively choose the model — or let Copilot choose via Auto mode. The multi-model capability surfaces in two places:
Copilot Cowork
Users can select GPT, Claude, or Auto before sending a prompt. Auto picks the model best suited to the task described — so users do not need to choose manually for every request.
Copilot Researcher
Two new capabilities: Critique (one model generates, a second reviews and refines before the final answer) and Model Council (submit one prompt and compare GPT vs Claude side by side in a split screen).
What Claude Opus 4.8 Is Good At
Claude Opus 4.8 is Anthropic's most capable model to date and is optimised for different strengths than GPT. Understanding this helps users (and IT admins briefing users) make sensible choices rather than just defaulting to whichever they recognise.
- Complex multi-step tasks and long-horizon work — Opus 4.8 maintains context and intent across very long workflows where simpler models lose track of earlier instructions
- Instruction following — it closely adheres to detailed, structured prompts and checks its own outputs before responding, reducing the need for follow-up corrections
- Document drafting and analysis — stronger than earlier models for drafting long-form documents, summarising complex material, and analysing data in Excel
- Coding tasks — particularly strong at complex, multi-step coding and debugging scenarios
- Image understanding — reads images at higher resolution than earlier Claude versions, useful for analysing diagrams, screenshots, and charts pasted into prompts
Practical guidance for users
For most everyday tasks — quick answers, summarising a meeting, drafting a short email — Auto mode is the right choice. For long research reports, complex document analysis, or multi-step projects, try Claude explicitly. For quick creative tasks or when you need a response that sounds distinctly "Microsoft 365", GPT remains solid. The Model Council side-by-side comparison is genuinely useful for high-stakes work where you want a second perspective.
What IT Admins Need to Know
There are three things worth being across before your users encounter Claude in Copilot:
EU Data Boundary exclusion
Anthropic models are currently excluded from the EU Data Boundary. Customers within the EU Data Boundary or in the UK have Anthropic models disabled by default. If your organisation is EU-based and you expect users to use Claude, check your tenant configuration and understand what this means for your data residency requirements.
No government cloud support yet
Anthropic models are not available in GCC, GCC High, or DoD environments. There is no FedRAMP certification in place yet. Government cloud tenants will not see Claude as an option.
Frontier Programme access
The Model Council and Critique features in Copilot Researcher are currently available to tenants with the Frontier Programme enabled. If your organisation has not enrolled, users will not see these capabilities yet. The broader Claude model choice in Cowork and Chat is rolling out to all Copilot tenants.
The bigger picture
Microsoft is positioning Copilot as model-agnostic — the best model for a given task, regardless of who built it. This strategy hedges against any single model provider and gives enterprise customers flexibility. For IT admins, it also means the governance conversation around AI in M365 is no longer just about OpenAI's data handling — you now need to understand Anthropic's subprocessor agreement with Microsoft too. The Anthropic models in Microsoft Online Services page on Microsoft Learn covers this.
Official Microsoft References
- Microsoft Tech Community — Available Today: Anthropic Claude Opus 4.8 in Microsoft 365 Copilot
- Microsoft Mechanics Blog — Claude + GPT: Multi-model Intelligence in Copilot
- Microsoft Mechanics YouTube — Claude + GPT | Multi-model Intelligence in Copilot (Video)
- Microsoft Learn — Anthropic Models in Microsoft Online Services
- Microsoft Learn — Choose a Model for Copilot Cowork