MAI-Voice-1 and MAI-1-Preview Set to Drive Microsoft’s Consumer Apps
,
Microsoft has taken a decisive step toward independence in the AI race with the unveiling of its first internally built models, signaling a shift away from its heavy reliance on OpenAI. The move reflects the company’s growing ambition to establish its own AI capabilities at scale and diversify its model portfolio for Copilot and beyond.
The two models announced this week are MAI-Voice-1 and MAI-1-preview, both developed entirely in-house by Microsoft AI. The announcement followed months of speculation about when the company would begin to showcase its own foundational work, and it now positions Microsoft as not just a partner of OpenAI but also a direct player in model development.
MAI-Voice-1 is designed for fast, high-quality speech generation, capable of producing one minute of audio in under a second on a single GPU. Microsoft has already integrated it into Copilot Daily, where it generates AI-powered news summaries, and into Copilot Podcasts, which allows users to create dynamic, prompt-based podcast episodes. A Copilot Labs demo further demonstrated its ability to deliver expressive, natural-sounding audio across both single and multi-speaker formats.
Alongside this, MAI-1-preview has been introduced for text-based use cases. It is currently being tested on LMArena, the model evaluation platform that has become a proving ground for AI newcomers. Microsoft is also offering developers early access through its API program, with plans to roll the model out within Copilot in the coming weeks. The company emphasized that user feedback will be central to refining MAI-1-preview’s performance before a wider release.