One-sentence summary
Answers common questions about Microsoft's first in-house models MAI-Voice-1 and MAI-1-preview launched November 1, 2026: positioning, training scale, where to try, and strategy.
FAQ
What is MAI-Voice-1?
It is Microsoft's ultra-fast speech generator, producing a minute of audio in under a second on a single GPU.
How much compute trained MAI-1-preview?
It was trained on about 15,000 Nvidia H100 GPUs, targeting everyday conversational tasks.
Where can I try it now?
Both are integrated into Copilot and open for public testing on Copilot Labs and LMArena.
Does this mean Microsoft no longer relies on OpenAI?
It is read as Microsoft building consumer-facing AI capability alongside its deep OpenAI partnership, reducing dependence on a single external provider, not fully severing ties.