Microsoft has begun using its own MAI models for some AI features in Excel and Outlook, according to Bloomberg Law and TechCrunch, reducing reliance on OpenAI and Anthropic for selected prompts while keeping third-party models in the mix.
Microsoft has begun replacing some third-party AI models with its own MAI models in parts of Microsoft 365, Bloomberg Law reported.
Bloomberg Law reported that Microsoft is using its in-house MAI models to handle some AI prompts in Excel and Outlook, replacing OpenAI and Anthropic models for certain workloads. The report said tens of thousands of weekly prompts are now being handled by Microsoft’s own models.
TechCrunch separately reported that Microsoft is relying more on its own models in Office apps as part of a broader effort to control AI costs. TechCrunch framed the move as part of an industry pattern in which companies route some tasks to smaller or cheaper models rather than using the most expensive frontier systems for every request.
The reported shift does not mean Microsoft is abandoning OpenAI or Anthropic models across Microsoft 365. The available reporting describes a partial substitution in selected applications and prompt types, not a wholesale replacement of third-party models throughout the Office suite.
Microsoft AI’s own announcement of seven new MAI models said the company designed the models for “first-party products.” In that announcement, Microsoft said one MAI model tuned for Excel could match GPT-5.4 on the company’s internal evaluation while operating at up to 10 times the efficiency.
That efficiency claim is Microsoft’s own benchmark, not an independent evaluation. Still, it helps explain why the company may use specialized internal models for high-volume tasks inside productivity software. If a narrower model can perform a defined Office task at lower inference cost, Microsoft has a financial incentive to use it where quality is sufficient.
The main business implication is cost management. Bloomberg Law’s report and TechCrunch’s follow-up both point to Microsoft reducing reliance on external model providers for some workloads. For a product like Microsoft 365 Copilot, where AI features can generate frequent user requests, inference costs matter directly to margins.
Microsoft also remains deeply tied to OpenAI through its investment, cloud relationship and product integrations. The reports do not support the claim that Microsoft is replacing OpenAI entirely, nor do they establish that all Excel, Outlook or Microsoft 365 AI features now run on MAI models.
A more precise reading is that Microsoft is building a model mix: using its own MAI systems for selected first-party productivity tasks, while continuing to use outside models where they are preferred or required.
For enterprise customers, the development shows that AI assistants inside productivity software may increasingly rely on multiple models behind the scenes. Users may not see which model answers a specific prompt, but vendors can route requests based on cost, speed, capability and product requirements.
For Microsoft, greater use of in-house models could reduce dependency on outside providers for routine Office workloads. For OpenAI and Anthropic, the reports suggest that large software platforms may become more selective about when they use frontier models, especially for repetitive or structured tasks.
The shift is significant, but the evidence supports a measured conclusion: Microsoft is moving some Microsoft 365 AI workloads to its own MAI models, not replacing all third-party AI across its productivity stack.
Microsoft has begun replacing some third party AI models with its own MAI models in parts of Microsoft 365, Bloomberg Law reported.
The report said tens of thousands of weekly prompts are now being handled by Microsoft’s own models.
TechCrunch separately reported that Microsoft is relying more on its own models in Office apps as part of a broader effort to control AI costs.
Continue reading
