Microsoft has announced the first artificial intelligence models fully designed within the company: MAI-VOICE-1 and MAI-1-PREVIEW.
Mai-Voice-1 is a voice model that can generate an audio for a minute on one video card in one second. It is already used in some Microsoft products, for example, in Copilot Daily, where AI voices the main news of the day, and to create podokast -like discussions.
MAI-1-Preview is created to work with the text: it can understand the instructions and give useful answers to everyday questions. Microsoft taught it for approximately 15,000 NVIDIA H100 graphic processors. The model is tested both in Copilot and on the platform for evaluating AI Lmarena.
The head of Microsoft AI Mustafa Suleiman noted that the company’s internal models are aimed at ordinary users, not corporate tasks.
