How Claude ended up inside Microsoft 365 Copilot
The September 2025 announcement — Claude joining Copilot alongside GPT, routed through AWS — was the prologue. The real pivot came in October 2025, when Microsoft and OpenAI restructured their contract. Microsoft won the right to develop its own models and partner with third-party labs. OpenAI won the right to buy compute from Oracle and AWS. The exclusivity clause was gone. Anthropic committed to buying roughly $30 billion of Azure capacity, and Satya Nadella made it very clear: Microsoft 365 Copilot would be multi-model by default, not by exception.
By January, that commitment hit production. Claude is now the default model for most tenants worldwide.
What actually shipped: the 7 January 2026 default switch
On 7 January 2026, Microsoft flipped Anthropic on as a subprocessor for Microsoft 365 Copilot across most commercial tenants worldwide. Claude Sonnet 4.5 (and, in Copilot Studio, Claude Opus 4.1) became accessible inside the Researcher agent, Copilot Chat, and agent-building flows. For customers in the EU, UK, and EFTA, admin opt-in is still required — Microsoft added the specific toggle in the M365 admin center on 3 April 2026.
This is not a pilot. It's the default for millions of seats. If your tenant is outside EU/UK/EFTA and nobody has changed a setting, Claude is already answering some of your colleagues' Copilot prompts.
The model selector: Claude Sonnet 4.5 vs GPT inside Copilot Chat
Inside Copilot Chat, licensed Microsoft 365 Copilot users now see a model selector at the top of the panel. Click the current model name and you can choose between OpenAI's GPT family and Claude Sonnet 4.5. The rollout of the selector completed in late March 2026.
The practical consequence is that "Copilot said X" is no longer a useful sentence. You need to know which model said X. Anecdotally, teams report Claude Sonnet 4.5 handling Excel formula generation, PowerPoint narrative structure, and long-document summarisation differently — often better — than default GPT. But the bigger point is that output variability is now a governance problem: two analysts asking Copilot the same question on the same file can get materially different answers depending on their model choice.
Copilot Cowork and the Microsoft 365 E7 tier
On 9 March 2026, Microsoft launched Copilot Cowork — an enterprise agent built with Anthropic's help, using the same agentic harness that powers Claude Cowork, but running inside a customer's Microsoft 365 tenant with enterprise data protection and a "Work IQ" layer drawn from the user's emails, files, meetings and chats.
Cowork is designed for long-running, multi-step marketing and ops work: "refresh our Q2 campaign brief across Word, pull the latest performance numbers from Excel, update the deck, and draft an email to the channel leads." It went broadly available in the Frontier program in late March 2026.
It is also the centrepiece of Microsoft 365 E7, the new enterprise SKU launching 1 May 2026 at $99 per user per month. That is roughly a 3x uplift on E5 for seats that need the full Cowork experience. For any marketing leader budgeting 2026 headcount, this is the single most material Microsoft pricing change of the year.
Claude in Microsoft Foundry: Azure as the multi-model cloud
Parallel to the Copilot story, Claude Sonnet 4.5, Haiku 4.5, and Opus 4.1 went live in Microsoft Foundry (formerly Azure AI Foundry) in early 2026. Claude Opus 4.6 followed, positioned for deep reasoning, coding and agent-driven workflows. Azure is now the only cloud offering both Claude and GPT frontier models to enterprise customers on a single platform.
For marketing and RevOps teams building custom agents, the implication is significant: your ISV partners no longer have to pick a model religion. They can ship a Foundry-based agent that picks Claude for long-context research and GPT for structured generation, on the same billing and governance stack.
The OpenAI contractual unwind — and why Microsoft needed a second horse
None of this is an accident. Three things pushed Microsoft to break exclusivity when they restructured their contract:
- Concentration risk. Running a $100B+ annualised productivity franchise on one external model provider is untenable — any outage, model regression, or safety incident at OpenAI would hit Copilot directly.
- Performance differentiation. Enterprise customers consistently reported that Claude handled certain long-document, structured reasoning and instruction-following tasks better. Losing those workloads to independent Claude deployments was a strategic leak.
- The Anthropic commitment. Anthropic's ~$30B Azure commitment wasn't just goodwill — it gave Microsoft a second frontier-lab anchor without the complications of the OpenAI equity structure.
The net result: Microsoft is now running a genuine multi-model strategy inside its flagship consumer and enterprise surface. Google is still shipping Gemini-only. OpenAI is shipping ChatGPT. Microsoft is shipping choice.
What this means for marketers in 2026
Three concrete actions before summer:
- Audit your Copilot tenant settings. If you're in EU/UK/EFTA, decide whether to opt in to Anthropic as a subprocessor — and get legal sign-off on data flow to Anthropic's infrastructure. If you're elsewhere, you're already defaulted in; document which model produced which artefact for any regulated output.
- Pilot Copilot Cowork before E7 pricing lands on your desk. The $99/user/month tier is a strategic fork. You need evidence — real tasks, real time savings — before your CFO gets the quote.
- Treat Copilot output as model-specific. Train your team to name the model in any prompt log or handoff. "Copilot wrote this" is no longer a sentence that tells you anything useful.
Most importantly, stop thinking of Copilot as a single AI. It is now a model marketplace wearing a Microsoft skin. The quality of your marketing output in 2026 depends on which model you route to which task — and nobody at Microsoft is going to make that decision for you. This is the same multi-model reality we explored when covering Anthropic's developer-tooling push — the frontier lab is showing up everywhere at once.
Anjin: the Marketing Operating System for a multi-model world
Anjin is the Marketing Operating System built for exactly this problem. We are a Marketing Operating System, not a Copilot skin. Anjin sits above the model layer — Claude, GPT, open-source, whatever comes next — and routes each step of a campaign to the model that's genuinely best at it. Research with long-context Claude. Structured JSON and tool use with GPT. Creative variation across both. One brief in, a finished campaign out, with the model choice logged for every asset.
We don't care which frontier lab wins this cycle. Neither should you. What marketers need is an operating system that absorbs the model chaos, enforces brand, produces output at the speed modern launches demand, and doesn't cost a full E7 upgrade per seat.
The £888 Lifetime License — Offer Closing Soon
Lifetime access to Anjin for a one-time payment of £888. Not a subscription. Not a seat. Not a trial. One payment, unlimited use, for as long as Anjin exists.
The average marketing team spends £888 in about three working days on tooling, freelancers and coordination software. You're buying the platform that replaces most of it — once.
This price will not be offered again once we close our early-access cohort.
Claim your £888 Anjin lifetime license →Founders, agency owners and in-house marketers — this is how you run marketing at AI speed without the team, the burn, or another year of waiting.
Sources: Microsoft 365 Blog, Microsoft Learn, Microsoft Support, UC Today, Microsoft Azure Blog, Azure Blog (Opus 4.6), Anthropic, Fortune, VentureBeat, Axios, Information Age, eWEEK
