Microsoft has released two new artificial intelligence models into public preview, saying they can drastically reduce costs compared to using OpenAI's technology. The models are MAI-Image-2.5-Pro, an image generator, and MAI-Voice-2-Flash, a speech model for business use.
According to Microsoft, using these models in its products has cut GPU costs by up to 89% in some cases. For example, in PowerPoint, the image model reduced costs by 84% compared to OpenAI's image model. In OneDrive, it boosted save rates by 26% and lowered latency. The speech model now powers Dynamics 365 Contact Center, used by companies like T-Mobile and EasyJet.
Microsoft said the models are already running in Bing, PowerPoint, OneDrive, Excel, and other products. The company called this a step toward using its own models instead of relying on OpenAI. CEO Satya Nadella said Microsoft can now deliver high-quality AI at lower cost for everyday tasks, while still using frontier models for advanced needs.
Critics noted that Microsoft's performance numbers come from its own tests, not independent benchmarks. But the company says the strategy is about making AI affordable for billions of users. The models are available in preview through Microsoft Foundry.