Microsoft's MAI models cut costs 85%, match frontier AI performance

Microsoft is rapidly expanding its in-house MAI family of AI models across key products including GitHub Copilot, Excel, and Outlook, according to CEO Satya Nadella's July 23, 2026 blog post. The strategy, dubbed 'Frontier Diffusion & Control,' focuses on optimizing AI systems for specific use cases to reduce inference costs while matching or exceeding the performance of frontier AI models from competitors like OpenAI. This move signals Microsoft's intent to diversify its AI stack and reduce reliance on external providers while maintaining enterprise-grade capabilities.

As part of this rollout, Microsoft released two new models: MAI-Image-2.5-Pro and MAI-Voice-2-Flash. The image model boasts being 85% cheaper than OpenAI's comparable offering, while the voice model has cut GPU costs by up to 89% in the Dynamics 365 Contact Center. These cost reductions are paired with performance claims that the models deliver similar or superior results to leading alternatives. By embedding MAI models directly into productivity tools and enterprise services, Microsoft aims to offer a more cost-effective, integrated AI experience for its users, potentially reshaping the competitive landscape of enterprise AI.