Microsoft is expanding its AI presence with the introduction of two new models developed and trained by its own teams. MAI–Voice-1 is the company’s first natural speech generation model, while MAI-1-preview is a text-based model and the first baseline model trained from start to finish.
Microsoft is expanding its AI presence with the introduction of two new models developed and trained by its own teams. MAI–Voice-1 is the company’s first natural speech generation model, while MAI-1-preview is a text-based model and the first baseline model trained from start to finish.
MAI–Voice-1 is currently being used in Copilot Daily and Podcast features. Microsoft has made MAI-1-preview available for public testing on LMArena and will begin previewing it in select Copilot scenarios in the coming weeks, Engadget reports .
According to Mustafa Suleiman, leader of Microsoft's AI division, both models were designed with an emphasis on efficiency and cost-effectiveness.
MAI–Voice-1 runs on a single GPU, while MAI-1-preview was trained on about 15,000 Nvidia H-100 GPUs. In comparison, other models, such as xAI’s Grok, required over 100,000 of these chips to train.
“The art of training models increasingly comes down to choosing the ideal data so as not to waste computing power on unnecessary tokens that don’t actually provide the model with significant knowledge,” Suleiman noted.
While Microsoft Copilot is used to test its own models, it is largely built on OpenAI’s GPT technology. The decision to build its own models, despite billions of dollars invested in the new AI company, suggests that Microsoft is aiming to become an independent player in this space.
Suleiman also noted that Microsoft has "a huge five-year plan that we're investing in quarter after quarter."
