2 Models Developed Internally at Microsoft Underscore Aggressive AI Ramp-Up, Hiring

As Microsoft continues aggressive moves to deliver and promote AI models from OpenAI — including the recently released GPT-5 — it’s also stepping up development of internally developed models, including two released late last week.

The company’s new models, under the Microsoft AI or MAI umbrella, include a natural speech generation model providing expressive audio, as well as a second model designed for consumers seeking to benefit from models that specialize in following instructions.

“We believe that orchestrating a range of specialized models serving different user intents and use cases will unlock immense value,” the company said in announcing  MAI-Voice-1 and MAI-1-Preview. The company indicated that customers should expect more models in the future as MAI pursues a mission of “creating applied AI as a platform for category defining and deeply trusted products that understand each of our unique needs.”

MAI-Voice-1 Details

MAI-1-Voice, the company’s first natural speech generation model, is available in Copilot Daily and Podcasts, and as a new Copilot Labs experience that customers can try today.  The company describes Copilot Labs as a hub for experimental AI initiatives. In Copilot Labs, customers can try expressive speech and storytelling demos.

MAI-Voice-1 delivers high-fidelity audio across single and multi-speaker conversations. It generates speech rapidly, creating a full minute of audio in under one second on a single GPU. MAI-Voice-1 currently powers Microsoft Copilot Daily and Podcasts features.

MAI-1-Preview Details

The company has begun public testing of MAI-1-Preview on LMArena, a popular platform for community model evaluation where AI can be evaluated through human preferences.

MAI-1-Preview is an in-house mixture-of-experts model, pre-trained and post-trained on approximately 15,000 NVIDIA GPUs. This model is designed for consumers seeking to benefit from models that specialize in following instructions and providing helpful responses to everyday queries. MAI-1-Preview is the first Microsoft foundation model trained end-to-end, and the company said that provides a solid view into future offerings inside Copilot.

MAI-1-Preview will be rolled out for text-based use cases within Copilot over the coming weeks with the goal of improving from user feedback. In so doing, the company will be able to evaluate millions of unique interactions to deliver optimal outcomes.

In addition to LMArena, MAI is making this model available to trusted testers, who can apply for API access.

Further Model Development

MAI underscored its commitment to developing and delivering new models in house with a call for developers to join Microsoft and contribute to its work. In an era when many companies are reducing their workforce because some functions are being managed by AI, a public solicitation for developers as part of a release of new models gives a clear indicator of the strategic importance of this work to the company and its customers, as well as the commitment to continue investing aggressively in AI. Software engineering appears to be a particularly rich category among those where Microsoft is hiring.

As part of the push for developers, the company emphasized its compute roadmap that’s supported by its operational NVIDIA GB200 cluster.

While the development of new models in-house will fuel speculation that Microsoft is mapping a future that is fully, or more, independent from OpenAI, the surge in AI usage and models — not only from Microsoft but from others including Google Cloud, which is also partnering with Oracle on AI models — suggests there’s room for more models to serve many needs, and that no one company can deliver on that fully or with a single partner.

---

For a 36-Hour Immersion into the FY27 Priorities that define Partner Success in the AI Era, join us at the AI Business Solutions Partner Executive Summit, running July 22-23, 2026, in Bellevue, Washington. Register today.