August 29, 2025 – Microsoft AI (MAI) announced two new in-house models designed to support its mission of building AI that empowers people and organizations worldwide.
The releases mark MAI’s next step in developing purpose-built models for Copilot and beyond:
1. MAI-Voice-1
A highly expressive, natural speech generation model.
Capable of producing a full minute of audio in under a second on a single GPU, making it one of the fastest speech systems available.
Already integrated into Copilot Daily and Podcasts, and available in Copilot Labs for demos such as interactive storytelling and guided meditations.
2. MAI-1-preview
MAI’s first end-to-end foundation model, built as a mixture-of-experts model and trained on ~15,000 NVIDIA H100 GPUs.
Optimized for instruction-following and useful everyday responses.
Currently being tested publicly on LMArena, with rollout planned for select text-based use cases in Copilot.
Also available to trusted testers via API access applications.
MAI emphasized its orchestration approach—leveraging a range of specialized in-house, partner, and open-source models to serve millions of unique interactions.
The team also highlighted its next-generation GB200 cluster, now operational, which will power further advances in model training and deployment.
Looking Ahead
Microsoft AI plans to expand its model lineup, refine its systems with user feedback, and bring its next generation of models into the hands of developers and enterprises globally.
Join the conversation! Your thoughts help the community grow.