Microsoft AI Makes a Big Move: Two In-House Models, MAI-Voice-1 and MAI-1-preview, Make a Stunning Debut

Microsoft AI (MAI) has unveiled its two latest powerful models: the ultra-efficient voice generation model MAI-Voice-1 and the large foundational model MAI-1-preview. This is not just a technological leap, but also a significant step in Microsoft’s commitment to creating AI for everyone and empowering every person on the planet. Let’s see how they will change the way people interact with AI.

At Microsoft AI (MAI), we hold a firm belief: AI should empower every person on the planet. The team is creating an AI companion that can serve all of humanity, providing support and assistance at any time. It will be a gateway to the universe of knowledge, offering a range of powerful capabilities to help individuals and organizations achieve more.

Microsoft AI’s goal is to build a responsible, reliable, and applied AI platform that is both personal and professional. This platform must not only define the future of the industry but also deeply understand the unique needs of each individual, becoming a trustworthy product. Since last year, the team has been focused on laying the groundwork for this vision, and today, Microsoft AI is showing the world the initial results of its efforts.

It is worth noting that both models introduced in this article are non-local models that run on cloud servers. Users will need an internet connection to access their powerful computing capabilities.

Hearing the Future? MAI-Voice-1 Brings Sound to Life

First up is MAI-Voice-1.

This is not just a voice model; it is Microsoft AI’s first voice generation model with high expressiveness and natural fluency. Voice is likely to be the primary interface for future AI companions, and MAI-Voice-1 was created for this purpose. It can provide high-fidelity, emotionally rich audio, effortlessly handling both monologues and multi-person dialogue scenarios.

Frankly, its efficiency is truly surprising. MAI-Voice-1 can generate a full minute of audio in less than a second on a single GPU, making it one of the most efficient voice systems available today.

Want to experience it for yourself? MAI-Voice-1 has been quietly launched in the Copilot Daily and Podcasts features. Not only that, but Microsoft AI has also opened a new experience area in Copilot Labs, allowing users to try out its powerful expressive and storytelling capabilities firsthand. Imagine creating a “choose-your-own-ending” adventure story with a simple prompt, or customizing a guided meditation to help you fall asleep peacefully.

More Than Just Conversation: MAI-1-preview Undergoes Public Testing

Next is Microsoft AI’s second major announcement: MAI-1-preview.

Microsoft AI has begun public testing of MAI-1-preview on LMArena, a well-known community model evaluation platform. This is not only MAI’s first fully end-to-end trained foundational model, but it also gives the outside world a glimpse of what the future of Copilot might look like.

MAI-1-preview is an in-house developed “mixture-of-experts” model. Simply put, it’s like having a group of specialists, each with their own expertise, work together to solve problems. It was pre-trained and post-trained on approximately 15,000 NVIDIA H100 GPUs and is designed to understand complex instructions and provide useful responses to everyday queries.

In the coming weeks, MAI-1-preview will be gradually applied to specific text-based use cases within Copilot, with the goal of learning and improving from user feedback. Of course, Microsoft AI will continue to use the best models from its team, partners, and the open-source community to enhance its products. This flexible strategy allows it to provide the best user experience across millions of unique interactions every day.

Additionally, API access to this model will be made available to trusted testers by application. The team is very excited to gather early feedback to understand where the model excels and how it can be made even better.

This Is Just the Beginning: Co-creating the Future of AI

Microsoft AI has ambitious goals for the future.

This announcement is just the beginning. The company believes that by integrating a series of specialized models tailored to different user intents and application scenarios, immense value can be unlocked.

MAI is a lean and fast-moving lab composed of top global talent, with an exciting roadmap for computing resources, and the new generation of GB200 clusters is now operational. More importantly, they have a grand mission they truly believe in. The team is fortunate to partner with exceptional product teams, giving its models the opportunity to reach billions of users and create a huge positive impact.

For talented, ambitious, and unconventional individuals, Microsoft AI continues to keep its doors open, inviting them to join in building the next generation of AI models.

More info: https://microsoft.ai/news/two-new-in-house-models/