Microsoft Unveils In-House AI Models: MAI-Voice-1 and MAI-1-Preview

Microsoft launches its first in-house AI models

Microsoft announced its first homegrown AI models on Thursday: MAI-Voice-1 AI and MAI-1-preview. The company says its new MAI-Voice-1 speech model can generate a minute's worth of audio in under one second on just one GPU, while MAI-1-preview "offers a glimpse of future offerings inside Copilot." Microsoft already uses MA1-Voice-1 to power a couple of its features, including Copilot Daily, which has an AI host recite the day's top news stories, and to generate podcast-style discussions to help explain topics. You can try MA1-Voice-1 out for yourself on Copilot Labs, where you can enter what you want the AI model to say, as well as change its voice and style of speaking. In addition to this model, Microsoft introduced MAI-1-preview, which it says it trained on around 15,000 Nvidia H100 GPUs. It's built for users in need of an AI model capable of following instructions and "providing helpful responses to everyday queries." Microsoft AI chief Mustafa Suleyman said during an episode of Decoder last year that the company's internal AI models aren't focused on enterprise use cases. "My logic is that we have to create something that works extremely well for the consumer and really optimize for our use case," Suleyman said. "So, we have vast amounts of very predictive and very useful data on the ad side, on consumer telemetry, and so on. My focus is on building models that really work for the consumer companion." The company plans on rolling out MAI-1-preview for certain text use cases in its Copilot AI assistant, which currently relies on OpenAI's large language models. It has also started publicly testing its MAI-1-preview model on the AI benchmarking platform LMArena. "We have big ambitions for where we go next," Microsoft writes in the blog post. "Not only will we pursue further advances here, but we believe that orchestrating a range of specialized models serving different user intents and use cases will unlock immense value."

Microsoft unveils home-made ML models amid OpenAI talks

Microsoft AI honcho insists partnership with Sam Altman's brainbox behemoth is alive and well Microsoft has introduced two home-grown machine learning models, potentially complicating negotiations with its current favored model supplier, OpenAI. On Thursday, Microsoft AI (MAI) debuted MAI-Voice-1, which generates realistic sounding speech from text, and MAI-1-preview, a model it's testing as the future basis of the company's Copilot service. In a Semafor video interview, MAI CEO Mustafa Suleyman explained Microsoft needs its own foundation models is because AI is fundamental to the company's business. "We have to be able to have the in-house expertise to create the strongest models in the world," he said. At the same time, Suleyman insisted Microsoft's collaboration with OpenAI has been successful so far, and expressed hope it will continue. If Redmond can create the strongest models in the world, it's unclear why it would continue paying OpenAI for less capable technology unless it's contractually obliged to do so. Microsoft has already invested around $13 billion in OpenAI and the two firms are reportedly trying to renegotiate their contract, set to expire in 2030, so that OpenAI can restructure for a future public offering. Separately, OpenAI is said to be discussing the potential sale of shares owned by employees to investors in a deal that would see the unprofitable firm valued at $500 billion. Last year, Microsoft opted not to release its VALL-E 2 speech synthesis project to the public because of potential abuses "such as spoofing voice identification or impersonating a specific speaker." OpenAI took similar steps, limiting access to its Voice Engine for speech synthesis. And when Consumer Reports looked at voice cloning services, it found most firms didn't do enough to prevent unauthorized impersonation. Yet MAI-Voice-1 has arrived in Copilot Labs with only a minimalist warning: "Copilot may make mistakes." It also powers Copilot Daily, an online AI-voiced summary of news and historic events, and Copilot Podcasts. "MAI-Voice-1 is a lightning-fast speech generation model, with an ability to generate a full minute of audio in under a second on a single GPU, making it one of the most efficient speech systems available today," said MAI in an online post. Microsoft let model evaluation platform LMArena test MAI-1, but it isn't available the public. Would-be testers in the US can apply for access. LMArena currently ranks Microsoft's model the equal thirteenth most effective in terms of output quality, behind grok-3-preview-02-24 and ahead of gemini-2.5-flash. "MAI-1-preview is an in-house mixture-of-experts model, pre-trained and post-trained on ~15,000 NVIDIA H100 GPUs," MAI said. "This model is designed to provide powerful capabilities to consumers seeking to benefit from models that specialize in following instructions and providing helpful responses to everyday queries." That's significantly fewer GPUs than the 100,000 Nvidia H100s powering xAI's Colossus supercomputer cluster. And it's comparable with Meta's Llama-3.1 model, which required over 16,000 Nvidia H100s. Microsoft, which says its GB200 cluster is now operational, expects to expose MAI-1-preview for specific Copilot scenarios in the coming weeks, so it can gather data about the model's performance.

Microsoft introduces a pair of in-house AI models

Microsoft is expanding its AI footprint with the of two new models that its teams trained completely in-house. MAI-Voice-1 is the tech major's first natural speech generation model, while MAI-1-preview is text-based and is the company's first foundation model trained end-to-end. MAI-Voice-1 is currently being used in the Copilot Daily and Podcast features. Microsoft has made MAI-1-preview available for public tests on LMArena, and will begin previewing it in select Copilot situations in the coming weeks. In an interview with , Microsoft AI division leader Mustafa Suleyman said the pair of models was developed with a focus on efficiency and cost-effectiveness. MAI-Voice-1 runs on a single GPU and MAI-1-preview was trained on about 15,000 Nvidia H-100 GPUs. For context, other models, such as xAI's Grok, took more than 100,000 of those chips for training. "Increasingly, the art and craft of training models is selecting the perfect data and not wasting any of your flops on unnecessary tokens that didn't actually teach your model very much," Suleyman said. Although it is being used to test the in-house models, Microsoft Copilot is primarily built on OpenAI's GPT tech. The decision to build its own models, despite having sunk in the newer AI company, indicates that Microsoft wants to be an independent competitor in this space. While that could take time to reach parity with the companies that have emerged as forerunners in AI development, Suleyman told Semafor that Microsoft has "an enormous five-year roadmap that we're investing in quarter after quarter." With some concerns arising that AI could be facing a bubble-pop, Microsoft's timeline will need to be aggressive to ensure taking the independent path is worthwhile.

Microsoft starts testing AI model that could escalate competition with OpenAI

Mustafa Suleyman, CEO of Microsoft AI, speaks at an event commemorating the 50th anniversary of the company at Microsoft headquarters in Redmond, Washington, on April 4, 2025. Microsoft has largely relied on OpenAI's artificial intelligence models to power AI features in its key products. It's trying to lessen that dependence. The software company said Thursday that it's begun publicly testing a homegrown AI model that could lead to enhancements to its Copilot assistant for consumers. The MAI-1-preview model is being tested on LMArena, a website where people can conduct evaluations. "We will be rolling MAI-1-preview out for certain text use cases within Copilot over the coming weeks to learn and improve from user feedback," Microsoft said in a blog post. The company has published a form where developers can request early access. Meanwhile, Microsoft remains a key backer of OpenAI and strategic partner to the AI startup that's now valued at about $500 billion. Microsoft has invested over $13 billion in OpenAI, which in turn relies on cloud infrastructure from Microsoft to run its models. Microsoft draws on models from OpenAI to power features in Bing, the Windows 11 operating system and other products. On LMArena, Microsoft's new model was ranked 13th for text workloads on Thursday, below models from Anthropic, DeepSeek, Google, Mistral, OpenAI and xAI. Microsoft said in its blog post that the model was refined with help from around 15,000 of Nvidia's H100 graphics processing units, and it also has a working cluster of Nvidia GB200 chips. "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," said Mustafa Suleyman, CEO of Microsoft's AI unit, in a post on X. Microsoft has previously developed small open-source language models under the name Phi. MAI-1-preview represents "our first foundation model trained end to end in house," Suleyman wrote on X. Suleyman used to compete with OpenAI at startup Inflection. Last year, Microsoft hired him and many of his Inflection colleagues. Before that, Suleyman was a co-founder of DeepMind, an AI research startup that Google bought in 2014. Suleyman's group inside Microsoft has been expanding, with about two dozen people coming from Google's DeepMind AI lab in recent months.

Microsoft Launches MAI-Voice-1 and MAI-1-Preview, Two In-House AI Models | AIM

Microsoft confirmed that trusted testers can apply for API access to the model. Microsoft AI has released its first in-house speech generation model, MAI-Voice-1, and begun public testing of its large language model MAI-1-preview, marking a step in the company's push to develop purpose-built AI systems. MAI-Voice-1 is already available in Copilot Daily and Podcasts, and is now accessible through Copilot Labs. According to Microsoft, the model can generate "a full minute of audio in under a second on a single GPU." Microsoft says it is designed to enable expressive, multi-speaker audio for interactive use cases such as storytelling and guided meditations. Alongside its speech model, MAI has started testing MAI-1-preview, an in-house mixture-of-experts model trained on about 15,000 NVIDIA H100 GPUs. The model is currently available for evaluation on LMArena,

Microsoft launches own AI models to take on OpenAI, Google - The Economic Times

Tech giant Microsoft has released its first in-house artificial intelligence (AI) models under the Microsoft AI (MAI) team. The two models are: Significance The Satya Nadella-led company will directly compete with the likes of Meta and OpenAI, the latter of which it has invested in since 2019. The MAI initiative was born out of the company's push to create homegrown models and cut its reliance on the ChatGPT maker, especially after its leadership overhaul and senior exits since late 2023. By taking the AI reins in its own hands, Microsoft gains more control over how its technology works and how much it costs. Microsoft has already deployed these models in its Copilot tools, including Copilot Daily and Copilot Labs. The new models are also designed to work better with Microsoft's products such as Windows, Office, and Teams. How do these stand out? Microsoft's models will mainly compete with OpenAI's GPT-4 and GPT-5, Google's Gemini, Anthropic's Claude and Meta's LLaMA. Each of these companies has a headstart in building and distributing its AI systems to consumers. Microsoft is counting on its vertical integration with everyday tools and its strong enterprise reputation to race ahead. The company also plans to build more specialised models and continue using a mix of its own, OpenAI's, and open-source models. This comes weeks after OpenAI officially launched GPT-5, terming it their most advanced model yet. The much-anticipated model features major upgrades in reasoning, coding, writing, health, and multimodal capabilities. Feedback from users, however, has been mixed. The model's capabilities were impressive said some, but the rollout itself faced issues. Many users complained about rate limits, touch-mode glitches, and the removal of older models, which the company had to reverse later.

Microsoft Launches Public Testing of First In-House Foundation Model | PYMNTS.com

By completing this form, you agree to receive marketing communications from PYMNTS and to the sharing of your information with our sponsor, if applicable, in accordance with our Privacy Policy and Terms and Conditions. The model, dubbed MAI-1-preview, is being tested on LMArena, a platform for community model evaluation, the company said in a Thursday (Aug. 28) blog post. "This represents MAI's first foundation model trained end-to-end and offers a glimpse of future offerings inside Copilot," the company said in the post. "We are actively spinning the flywheel to deliver improved models." MAI-1-preview is designed for use by consumers and specializes in following instructions and answering everyday questions, according to the post. It will be rolled out for some text use cases in Copilot in the coming weeks, per the post. "We will 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," MAI said in the post. "This approach gives us the flexibility to deliver the best outcomes across millions of unique interactions every day." CNBC reported Thursday that Microsoft powers the artificial intelligence features of its Bing search engine, its Windows 11 operating system and other products primarily with AI models from OpenAI -- a company in which Microsoft has invested over $13 billion -- and that the development of an in-house model could signal that it's working to reduce that dependence. Microsoft added OpenAI to a list of competitors in its annual report last year, while OpenAI has added cloud providers beyond Microsoft, including CoreWeave, Google and Oracle, according to the report. MAI also announced in its Thursday blog post that it is releasing a natural speech generation model called MAI-Voice-1, making it available in Copilot Daily and Podcasts and as a Copilot Labs experience. "Voice is the interface of the future for AI companions and MAI-Voice-1 delivers high-fidelity, expressive audio across both single and multi-speaker scenarios," MAI said in the post. This announcement came on the same day that OpenAI released what it calls its "most advanced speech-to-speech model yet." The company also made its application programming interface Realtime API generally available, saying the application programming interface now has features that help developers build voice agents.

Microsoft breaks free from OpenAI reliance, launches two homegrown AI models

MAI-Voice-1 is said to produce a minute of audio in less than a second on a single GPU. Microsoft is stepping up its AI game with the launch of two homegrown models: MAI-Voice-1 AI and MAI-1-preview. These new AI models show the company's ambitions to create its own AI technology instead of relying solely on OpenAI's models. According to Microsoft, the MAI-Voice-1 is a "lightning-fast speech generation model." It is said to produce a minute of audio in less than a second on a single GPU. This model is already powering some of Microsoft's features, including Copilot Daily and Podcasts features. Users can try out MAI-Voice-1 in Copilot and Copilot Labs. "Voice is the interface of the future for AI companions and MAI-Voice-1 delivers high-fidelity, expressive audio across both single and multi-speaker scenarios," the company claims in a blogpost. Also read: Govt's online gaming ban lands in court as Indian firm mounts legal challenge Alongside MAI-Voice-1, Microsoft introduced MAI-1-preview. This AI model was pre-trained and post-trained on roughly 15,000 Nvidia H100 GPUs. It is designed to follow instructions and provide helpful responses for everyday queries. The company has begun public testing of MAI-1-preview on LMArena, a popular platform for community model evaluation. Also, Microsoft will be rolling MAI-1-preview out for certain text use cases within Copilot over the coming weeks to learn and improve the model. "We have big ambitions for where we go next," Microsoft said. "Not only will we pursue further advances here, but we believe that orchestrating a range of specialized models serving different user intents and use cases will unlock immense value." Also read: Samsung Galaxy event on Sept 4: Galaxy S25 FE, Tab S11 series and more expected With these new models, Microsoft is signaling that it wants to expand its AI ecosystem with models built in-house. While Copilot currently depends on OpenAI's large language models, the company's new offerings could eventually reduce that dependency.