(Bloomberg) — Looking to reduce AI costs, Microsoft is starting to replace OpenAI and Anthropic with its own models in software products like Excel and Outlook.
Most Read Articles on Bloomberg
Tens of thousands of AI prompts in widely used spreadsheet and email applications are now completed weekly using MAI models built in-house by Microsoft, according to people familiar with the effort. Previously, Excel and Outlook relied heavily on models from OpenAI and Anthropic, said the person, who was granted anonymity to discuss internal matters.
The scale of the MAI model’s use in Microsoft’s workplace products has not previously been reported. A Microsoft spokesperson declined to comment.
Although AI remains a small portion of overall usage, the move shows Microsoft is making progress in its efforts to build competitive artificial intelligence models at low cost. Mustafa Suleiman, head of AI models, said in June that the company was looking to reduce spending on Anthropic by using more MAI models.
Microsoft uses large amounts of AI tokens, the unit of consumption for AI computing, in products such as its workplace assistant Copilot. For now, the company can get much of its technology at a discount thanks to its long-standing partnership with OpenAI. But the clock is ticking on that deal, and Suleiman’s team is working to ensure that Microsoft doesn’t end up paying what major AI labs charge.
The company announced seven new AI models at its annual Build conference for developers in June. This includes being able to match the coding capabilities of the previous generation Anthropic model (Opus 4.6), which remains very popular, at a lower cost. “We pay a lot of money to Anthropic, so our goal is to reduce and eventually eliminate that cost,” Suleiman said at the time.
Microsoft’s MAI model is also available within GitHub Copilot, the company’s service for AI-assisted software development. Suleiman said the transcription model built by Microsoft will start being used in the Teams video conferencing app and other products in the coming months.
–With assistance from Matt Day.
Most Read Articles on Bloomberg Businessweek
