Microsoft put two in-house models into public preview, MAI-Image-2.5-Pro for high-fidelity image generation and MAI-Voice-2-Flash for high-volume speech, alongside production data claiming steep cost savings over OpenAI models. Microsoft says MAI-Image-2.5 cuts GPU costs up to 84% in PowerPoint compared with OpenAI's GPT-Image-2, while MAI-Voice-2-Flash cuts them up to 89% in Dynamics 365 Contact Center, whose customers include T-Mobile and EasyJet. Bing Image Creator now runs entirely on MAI. Satya Nadella wrote that Microsoft is starting to route traffic across its own surfaces to MAI whenever those models match or beat frontier alternatives, with OpenAI and Anthropic models staying in the orchestration for frontier work. All the cost and performance figures come from Microsoft's internal evaluations rather than independent benchmarks.
