Microsoft AI Releases Its First AI Models with Some Hints for the Future

A year and a half ago, Microsoft created a new Microsoft AI organization to create purpose-built AI models in-house and lessen its reliance on OpenAI. Today, we’re finally getting our first peek at that organization’s work, thanks to the release of its first two models, one straightforward and one quite mysterious.

“At Microsoft AI (MAI) we believe AI should be used to empower every person on the planet,” the announcement post explains. “We are creating AI for everyone, a supportive, helpful presence always in the service of humanity. It will be the gateway to a universe of knowledge and a set of capabilities that enable people and organizations to achieve more. Responsible, reliable, filled with personality and expertise, we are focused on creating applied AI as a platform for category defining and deeply trusted products that understand each of our unique needs.”

Microsoft AI was the first major and visible change at the company in the wake of the late 2023 Sam Altman drama at OpenAI, a wake-up call for the software giant that made it realize it needed an exit strategy. DeepMind and Inflection co-founder Mustafa Suleyman joined Microsoft to lead this organization, and he was joined by Inflection chief scientist Karén Simonyan and, we later learned, most of the team from Inflection.

Microsoft AI was odd from the get-go: Microsoft never explicitly promoted it as a Plan B of sorts should OpenAI implode, but it was widely understood that was the point. And at the time it began, Microsoft CEO Satya Nadella only told employees that Microsoft AI would “focus on advancing Copilot and [Microsoft’s] other consumer AI products and research,” with Suleyman reporting directly to him. He said that Microsoft would continue to build on its OpenAI infrastructure, build out its infrastructure to support OpenAI, and also “innovate and build products on top of their foundation models.” How the Microsoft AI announcement “reinforced” Microsoft’s “partnership construct and principles” was unclear.

Today, however, Microsoft AI announced its first two original AI models. And in doing so, it has provided more hints about its plans for this organization and a post-OpenAI world.

The first of the two models is straightforward. MAI-Voice-1 is a “highly expressive and natural speech generation model,” and it’s available now in Copilot Daily, Copilot Podcasts, and Copilot Labs. And it will be rolling out in Copilot in the coming weeks to address “certain text use cases.” This is clearly the first step in lessening Copilot’s (and thus Microsoft’s) reliance on OpenAI.

“Voice is the interface of the future for AI companions,” Microsoft AI explains, driving home a contentious point for old-school Windows users in recent weeks. “And MAI-Voice-1 delivers high-fidelity, expressive audio across both single and multi-speaker scenarios.”

The second model is more mysterious. It’s called MAI-1-preview, it’s available now on LMArena, a platform for evaluating community models, and it’s coming soon to “trusted testers” via an API.

“MAI-1-preview represents MAI’s first foundation model trained end-to-end and offers a glimpse of future offerings inside Copilot,” Microsoft AI hints. “We are actively spinning the flywheel to deliver improved models [and] we’ll have much more to share in the coming months.” By which I assume they mean Ignite in November.

According to this post, Microsoft AI is specifically creating a range of highly specialized AI models that will “serve different user intents and use cases, unlocking immense value.” The goal for each of these models is that they be the best possible solution for their respective workload.

“We continue to use the very best models from our team, our partners, and the latest innovations from the open-source community to power our products,” Microsoft AI says, finally spelling out how this transition is unfolding. “This approach gives us the flexibility to deliver the best outcomes across millions of unique interactions every day.”