Microsoft’s New AI Models Add More Naturalistic Speech to Copilot – and Rival OpenAI

執筆者
Megan Crouse
Megan Crouse
Aug 29, 2025
2 minute read
Microsoft releases MAI-1-preview, a foundation model that is undergoing public testing.

Source: Microsoft AI Copilot

eWeek のコンテンツおよび製品のおすすめは、編集上の独立性を保っています。パートナーへのリンクをクリックすると、当社が報酬を得る場合があります。 詳細を見る

Microsoft has pulled back the curtain on two proprietary AI models, the first time Redmond has developed direct competitors to its partner OpenAI. MAI-Voice 1 is a speech-generation model that users can try out now in Copilot Daily, Podcasts, and Copilot Labs. MAI-1-preview, a foundation model, is undergoing public testing in the community model evaluation forum LMArena. 

What do MAI-Voice 1 and MAI-1-preview do?  

MAI-Voice 1 provides a naturalistic, high-quality synthetic voice, Microsoft said. The company also touted the model’s efficiency: it can produce a minute of audio using only a single GPU. In Copilot Labs, users can generate audio from prompts. For example, Microsoft recommends asking it for a story or a guided meditation.  

Copilot Daily generates a 40-second summary of top headlines, while Copilot Podcasts creates natural-sounding conversations about articles or topics. Users can interact with the AI podcasters to steer the conversation or ask follow-up questions. 

MAI-1-preview enters community testing

MAI-1-preview is more of a generalist, a mixture-of-experts model that can answer day-to-day prompts. Microsoft plans to gather user feedback on certain uses for MAI-1-preview’s text generation within Copilot starting within weeks, the company said.

Testers can apply for early API access to MAI-1-preview. 

Both are primarily focused on consumer use, not enterprise use. 

Why is Microsoft making its own generative AI models?

Historically, Microsoft has provided much of the cloud infrastructure used by OpenAI, while Sam Altman’s company focuses on creating new AI models. The Microsoft AI products benefit from the stable of infrastructure. 

MAI-1-preview was trained in-house on approximately 15,000 NVIDIA H100 GPUs, Microsoft stated. At the same time, the next-generation GB200 cluster of Nvidia chips joined Microsoft’s stable, enabling further compute for AI efforts.

“We have big ambitions for where we go next — model advancements, an exciting roadmap of compute, and the chance to reach billions of people through Microsoft’s products,” Microsoft AI CEO Mustafa Suleyman wrote on X on Aug. 28. “We’re building AI for everyone.” 

Advertisement

Microsoft and OpenAI’s close partnership may begin to include some competition. Selling in-house models puts Microsoft in competition with it as well as Anthropic, DeepSeek, Meta, xAI, and others. 

Nvidia and the US government agreed to a 15% levy on sales of AI chips to China, but Nvidia says regulatory documents haven’t been finalized. 

Megan Crouse

Megan Crouse has a decade of experience in business-to-business news and feature writing, including as first a writer and then the editor of Manufacturing.net. Her news and feature stories have appeared in Military & Aerospace Electronics, Fierce Wireless, TechRepublic, and eWeek. She copyedited cybersecurity news and features at Security Intelligence. She holds a degree in English Literature and minored in Creative Writing at Fairleigh Dickinson University.

eWeek Logo

eWeek has the latest technology news and analysis, buying guides, and product reviews for IT professionals and technology buyers. The site's focus is on innovative solutions and covering in-depth technical content. eWeek stays on the cutting edge of technology news and IT trends through interviews and expert analysis. Gain insight from top innovators and thought leaders in the fields of IT, business, enterprise software, startups, and more.

TechnologyAdvice が所有・運営しています。 © 2026 TechnologyAdvice. 無断転載を禁じます

広告主に関する開示:このサイトに掲載されている製品の一部は、TechnologyAdvice が報酬を受け取っている企業のものです。この報酬は、製品がこのサイトのどこにどのように表示されるか(表示される順序など)に影響する場合があります。TechnologyAdvice は、市場で入手可能なすべての企業やすべての種類の製品を掲載しているわけではありません。