Microsoft released two new in-house AI models into public preview on July 23, publishing production data that shows its homegrown models cut GPU computing costs by up to 89 percent compared to OpenAI alternatives. MAI-Image-2.5-Pro targets premium image generation with precise text rendering, while MAI-Voice-2-Flash handles high-volume enterprise speech at twice the speed and 32 percent lower cost than its predecessor.
The announcement from Microsoft AI’s Superintelligence team reveals that Bing Image Creator, PowerPoint, OneDrive, Dynamics 365, Excel, and GitHub Copilot now run on Microsoft’s own models at least in part. The company said in its blog that each enhancement is a step toward the same goal: Microsoft products powered by Microsoft models.
MAI-Image-2.5 is already ranked number two for image editing on the Arena community leaderboard. The Pro variant costs $5 per million text input tokens and $106 per million image output tokens. MAI-Voice-2-Flash is priced at $15 per million characters, targeting call centers and real-time voice agents where latency and cost per call matter most.
