What AI model does it use?

What AI model does Microsoft Copilot use?

There are two Copilots worth separating before anything else. The free consumer app at copilot.microsoft.com is one. The other is the work Copilot inside Word, Excel, Outlook and Teams, which Microsoft has just renamed: a note on its Learn documentation reads "Microsoft 365 Copilot is now named Microsoft Copilot, and Microsoft 365 Copilot Chat is now named Microsoft Copilot Chat." So when Microsoft's own pages say "Microsoft Copilot" in September 2026, they usually mean the work one.

The work Copilot: GPT and Claude side by side

On September 22, 2026, Microsoft's Xia Song wrote that "Anthropic's Claude Opus 5.5 and OpenAI's GPT-6 Sol are rolling out to Microsoft Copilot today across Word, Excel, PowerPoint, Chat, Cowork, and Copilot Studio." That one sentence is most of the answer. Copilot isn't an OpenAI-only product any more.

Claude's place in it is bigger than I'd assumed. Microsoft's docs say it "enables Anthropic models on by default for most customers in commercial cloud (excluding EU/EFTA and UK)." Claude became a selectable option in Copilot Chat in June 2026, and in Word, Claude Sonnet 5 is the default for higher-reasoning tasks, according to the release notes. If you want to know when it's Claude answering, Microsoft says the web, desktop and mobile apps show an indicator.

Auto, and the models that cost extra

Microsoft announced new Copilot licensing on September 25, 2026, the day I checked. The seat license covers Copilot in the Office apps plus model selection, and "Auto is at the heart" of it: "Auto weighs accuracy, speed and cost on each request to route to the model best suited for the job."

The biggest models sit outside that seat. The same post says the new long-running agent features "and frontier models like Astra and Fable all run on UBB", Microsoft's usage-based billing. Astra is OpenAI's GPT-6 Astra and Fable is Anthropic's Claude Fable, so the top model from each lab costs extra per use. Admins can also limit which model families a group of users sees, which limits what Auto can pick for them.

The consumer Copilot app

The free app is harder to pin down. The newest first-party statement I could find is Microsoft's support page on conversation modes, which still says "Smart: Use the GPT-5 model to respond." That page names no newer model, and I found no 2026 announcement that replaces it. So for the consumer app, the honest answer is OpenAI's GPT-5 in Smart mode, as far as Microsoft has said, with no word on whether it has moved on.

Where Microsoft's own models fit

Microsoft builds models too, under the MAI name. Its release notes put MAI-Image-2-Efficient in PowerPoint for image generation, and MAI-Code-1.1-Flash is in GitHub Copilot (a separate product). Microsoft AI launched a reasoning model, MAI-Thinking-1, on August 12, 2026, saying it "was trained without distillation from third party models". But the announcement doesn't say it powers Copilot chat, and I didn't find any Microsoft page that does.

What Microsoft doesn't say

Microsoft doesn't publish which model Auto sends a given request to, or how often it chooses Claude over GPT. For the consumer app, it hasn't said anything newer than GPT-5.