Microsoft has largely relied on OpenAI 's models to power its AI products, but now it's working on its own in-house models.

On Thursday, Microsoft launched a speech generation model called MAI-Voice-1, and started public testing for a foundation model called MAI-1-preview. The company said MAI-Voice-1 can generate a minute of audio in under a second on a single GPU, highlighting its efficiency. MAI-Voice-1 is already powering Microsoft's Copilot Daily and Podcast features which provide AI-generated audio news recaps and personalized podcast-style content, much like Google's NotebookLM .

MAI-1-preview is currently restricted to LMArena, where users can try it out in head-to-head comparisons against other models and to trusted testers through the API. Microsoft says it will begin ro

See Full Page