Microsoft is Reportedly Transitioning to In-House AI Models

Bloomberg reported today that Microsoft has started transitioning away from OpenAI and Anthropic AI models in an attempt to cut costs. Their replacements? Microsoft AI (MAI) models, despite repeated assertions that its in-house models aren’t as sophisticated as the top external frontier models.

Still, this makes some sense given the use cases: Microsoft is starting to use MAI models within Microsoft 365 apps like Excel and Outlook, where the needs are more finite and readily understood. And though MAI model usage internally is still small, overall, according to Bloomberg, the software giant is making progress.

Microsoft announced several new MAI models at Build a few months back, including its first-ever flagship reasoning model, and there were hints at that time that Microsoft was starting to transition away from OpenAI and Anthropic. For example, the new MAI-Image-2.5 was already being used by PowerPoint, another Microsoft 365 app. MAI models are also used with GitHub Copilot, Microsoft’s developer solution.

And Microsoft AI CEO Mustafa Suleyman has always been clear about his team’s goal of eliminating the company’s reliance on outside models. Among other things, he said recently that an MAI transcription model, MAI-Transcribe-1.5, would soon be used by Microsoft Teams, inarguably one of the most important Microsoft 365 apps.

So we’ll see. This feels a little early, but this transition was inevitable.