Microsoft is shifting AI model usage in Copilot products like Excel and Outlook from OpenAI and Anthropic to its in-house MAI models. Bloomberg reports that these in-house models are already handling tens of thousands of requests per week in these applications, which previously relied more heavily on external models. The transition aims to lower Microsoft's costs on third-party AI services over time.
The MAI models are also being integrated into GitHub Copilot, with a proprietary transcription model set to launch in Teams. At the Build conference, Microsoft unveiled seven new AI models, including MAI-Thinking 1, its first reasoning model. Microsoft claimed it could match Sonnet 4.6 and Opus 4.6 in coding based on human evaluations, but benchmark results showed it lagged significantly behind competitors, performing roughly on par with Deepseek V3.2.
Microsoft's head of AI, Mustafa Suleyman, openly acknowledged the cost-cutting plan in June, stating, 'We pay a lot of money to Anthropic—so our goal is to reduce and ultimately eliminate that cost.' This strategy aligns with Microsoft's broader push for platform neutrality and potential shifts toward usage-based AI billing.
Source: thedecoder
