Microsoft reveals two in-house AI models: MAI-Voice-1 and MAI-1-preview

Microsoft has unveiled its first major in-house AI models, MAI-Voice-1 and MAI-1-preview, signaling a strategic move to reduce its dependence on OpenAI.

Until now, Microsoft had been largely dependent on OpenAI's AI models. Whenever OpenAI unveiled its newest models, Microsoft announced day-one availability of those models on Azure and across its products and services. Today, Microsoft announced two new in-house AI models: MAI-Voice-1 and MAI-1-preview.

MAI-Voice-1 is a speech generation model and is already available in Copilot Daily and Podcasts. To preview the full capabilities of this voice model, Microsoft has created a new Copilot Labs experience that anyone can try today.

With the Copilot Audio Expressions experience, users can just paste text content and select the voice, style, and mode to generate high-fidelity, expressive audio. They can also download the generated audio if required. Microsoft also highlighted that this MAI-Voice-1 model is very fast and efficient. In fact, it can generate a full minute of audio in under a second on a single GPU.

Second, Microsoft has begun public testing of MAI-1-preview on LMArena, a popular platform for community model evaluation. This represents MAI’s first foundation model trained end-to-end and offers a glimpse of future offerings inside Copilot. They are actively spinning the flywheel to deliver improved models and will have much more to share in the coming months.

MAI-1-preview is an MoE (mixture-of-experts) model, pre-trained and post-trained on nearly 15,000 NVIDIA H100 GPUs. Notably, MAI-1-preview is Microsoft's first foundation model trained end-to-end in-house. Microsoft claims that this model is better at following instructions and can offer helpful responses to everyday user questions. Microsoft will be rolling out this new model to certain text use cases within Copilot over the coming weeks.

Microsoft reiterated that MAI-1-preview is not replacing OpenAI's models in Copilot. Microsoft's plan is to use the best models from its own team, its partners, and even the open-source community. Microsoft has also released MAI-1-preview in LMArena to test its capabilities. In addition, this new text model is available via API for trusted testers.

The Microsoft AI team wrote the following regarding the launch of their new models:

We have an exciting roadmap of compute at MAI, with our next-generation GB200 cluster now operational. And we have an ambitious mission we truly believe in. We’re also fortunate to partner with incredible product teams giving our models the chance to reach billions of users and create immense positive impact.

By developing its own foundational models alongside leveraging those from OpenAI, Microsoft is building a more resilient AI foundation for its products and services.