Microsoft launches new AI models MAI-Voice-1 and MAI-1 Preview
MAI-1 Preview, trained on 15,000 NVIDIA H100 GPUs, enhances instruction-following and complex task performance.

Microsoft has unveiled two new AI models, marking a major step in its efforts to build its own technology rather than rely solely on OpenAI.
The first model, MAI-Voice-1, generates high-fidelity audio and supports both single and multi-speaker scenarios. Microsoft said the system can create a full minute of expressive audio in under a second on a single GPU, making it one of the fastest of its kind.
MAI-Voice-1 is already available in Copilot Daily and Podcasts, while Copilot Labs allows users to experiment with storytelling and speech demos. Microsoft sees voice as a vital interface for future AI companions.
MAI-1 Preview is currently undergoing community testing on LMArena and will soon be integrated into selected Copilot use cases. Microsoft said it plans to expand its family of specialised models, aiming to orchestrate different systems for diverse user needs.
Would you like to learn more about AI, tech and digital diplomacy? If so, ask our Diplo chatbot!