Microsoft has reportedly begun routing some Microsoft 365 AI prompts to its own MAI models, reducing reliance on OpenAI and Anthropic for select Office tasks while continuing to promote model choice across Copilot.
TL;DR
- Microsoft is reportedly using in-house MAI models for some Microsoft 365 prompts.
- Microsoft has not publicly confirmed the app-level routing change.
- The confirmed part is Microsoft’s wider push into MAI models and model choice.
- OpenAI and Anthropic remain part of Microsoft 365 Copilot’s AI ecosystem.
Microsoft has reportedly begun using its own in-house artificial intelligence (AI) models to answer a portion of prompts inside Microsoft 365 apps, marking another sign that major tech companies are looking to manage the rising cost of AI services.
The report, first carried by Bloomberg and cited by TechCrunch, says Microsoft is relying more on its homegrown MAI models and less on models from OpenAI and Anthropic for some Office-related AI tasks. However, Microsoft has not publicly confirmed the exact app-level implementation. TechCrunch said Microsoft had nothing further to share when contacted.
TechCrunch said the reported routing is applied to Excel and Word. As such, the safer verified framing is that the shift reportedly affects some Microsoft 365 apps, including Excel, rather than naming every affected app as confirmed.
The move does not mean Microsoft has stopped using third-party models. In September 2025, Microsoft said Copilot would continue to be powered by OpenAI’s latest models, while also adding Anthropic models to Microsoft 365 Copilot, starting with Researcher and Copilot Studio.
The reported change also comes after Microsoft AI made a stronger push into its own model family. Microsoft AI CEO Mustafa Suleyman announced “a family of seven new models developed in-house at Microsoft AI.” The lineup includes MAI-Thinking-1, MAI-Code-1-Flash, MAI-Image-2.5, MAI-Image-2.5-Flash, MAI-Transcribe-1.5, MAI-Voice-2, and MAI-Voice-2-Flash.
Microsoft said these models span reasoning, coding, image generation, transcription, and voice. It also said MAI-Code-1-Flash is designed for GitHub Copilot, VS Code, and the Microsoft stack, while MAI-Transcribe-1.5 supports domain-specific terminology across 43 languages.
The cost angle is important. Microsoft has described some MAI models in terms of price-performance and efficiency, including MAI-Code-1-Flash as comparable to Haiku but cheaper. Microsoft also said MAI-Image-2 starts at $5 per 1 million text-input tokens and $33 per 1 million image-output tokens.
The broader enterprise AI market is moving in a similar direction. Reuters reported that Microsoft launched Microsoft Frontier Company with $2.5 billion in funding to help customers select and integrate AI tools from Microsoft and outside providers. Judson Althoff, CEO of Microsoft Commercial Business, told Reuters, “we made a mistake by binding it to OpenAI models only.”
For Microsoft, the reported Office routing shift fits into a larger strategy: keep OpenAI and Anthropic available where needed, while using MAI models where Microsoft can control performance, cost, and product integration more directly.
For users, the practical impact is not yet clear. Microsoft has not disclosed how many prompts are being routed to MAI models, which Microsoft 365 apps are fully affected, or whether users will see any visible change in Copilot responses.
