Microsoft Quietly Reroutes Office AI Prompts to In-House Model, Asks $13 Billion Strategic Partner to Please Not Read Into It

REDMOND— Microsoft has begun replacing OpenAI and Anthropic models inside Excel and Outlook with its own MAI family of in-house models, completing what the company described as a routine infrastructure efficiency initiative inside a strategic partnership the company simultaneously described as "the most important in technology."

The shift, first reported by Bloomberg, has moved tens of thousands of weekly prompts from GPT-class and Claude models to Microsoft's MAI-Thinking-1 and MAI-Code-1-Flash, which the company says match incumbent performance for standard Office tasks at substantially lower inference cost. The change does not affect the partnership's structure, Microsoft said, as the partnership's structure does not specify which model handles email autocomplete.

Microsoft has invested more than $13 billion in OpenAI. OpenAI runs on Microsoft Azure. The two companies have appeared together at dozens of press events, shared a stage at multiple product launches, and in 2023 described their arrangement in terms that neither party's communications team has formally updated.

According to a person familiar with the situation, relevant OpenAI personnel learned of the routing change through a Bloomberg article. A Microsoft 365 Copilot summary of that article, processed by MAI-Thinking-1, was delivered to the relevant OpenAI executives. The summary noted that Microsoft had "identified cost efficiencies," recommended three action items, and flagged the article's overall sentiment as "mixed."

Microsoft confirmed the account's accuracy and said it reflects how the partnership is "evolving."

"We are deeply committed to our relationship with OpenAI," said a Microsoft spokesperson. "We just find it works better when Sam isn't doing the spreadsheets."

"We appreciate that," said an OpenAI spokesperson.