Microsoft deploys in-house MAI models across production products, claims up to 89% GPU cost reduction versus OpenAI equivalents

Microsoft released MAI-Image-2.5-Pro and MAI-Voice-2-Flash into public preview and disclosed that its internally built MAI model family now runs in production across Bing, PowerPoint, OneDrive, Dynamics 365, Excel, GitHub Copilot, and Azure, reporting GPU cost reductions of 84% in PowerPoint and 89% in Dynamics 365 Contact Center versus equivalent OpenAI models. CEO Satya Nadella framed the shift as deliberate traffic routing toward in-house models "whenever our models match or outperform frontier alternatives," explicitly positioning OpenAI and Anthropic models as interchangeable components in an orchestration layer rather than default infrastructure. The company also disclosed that a small coding model fine-tuned inside an Excel reinforcement learning environment reaches parity with GPT-5.6 on common spreadsheet tasks while running on H100 and A100 hardware, reducing dependence on latest-generation accelerators. For financial institutions, the disclosures structurally reframe the Microsoft-OpenAI revenue relationship and provide the most detailed self-reported evidence to date that large enterprise deployments can substitute task-specific models for frontier models at significantly lower unit cost.