Microsoft just shipped its own AI models, and they’re coming to OpenAI and Google. The company has publicly released three proprietary models: MAI-Transcribe-1, MAI-Voice-1 and MAI-Image-2. The models are available via the Microsoft Foundry platform and the MAI Playground.
So what can Microsoft’s AI models actually do?
The trio covers a variety of use cases: listening, speaking and seeing. MAI-Transcribe-1, for example, processes speech to text in 25 languages and is 2.5 times faster than Microsoft’s own Azure Fast offering. It is worth mentioning that the audio model was built by a team of 10 people.
MAI-Voice-1 can produce 60 seconds of natural-sounding audio in just one second. It also supports creating custom voices from just a short audio clip. MAI-Image-2 has already secured a place in the top three of the Arena.ai image generation rankings. Rollouts are currently underway in Bing and PowerPoint.
However, none of this happened overnight. The company has been contractually prohibited by none other than OpenAI from developing its own borderline AI until October 2025. Both companies signed a deal in 2019 that gave Microsoft a license to OpenAI’s models in exchange for providing OpenAI’s cloud infrastructure.
Is Microsoft ready to cut ties with OpenAI?
However, the deal also prohibited Microsoft from developing its own AI models. When that changed, Microsoft released its own AI models that power Copilot and Teams in the background. The models are available for any developer to build on Foundry.
Not yet. Microsoft AI CEO Mustafa Suleyman has reiterated the company’s commitment to its OpenAI partnership, even as these models suggest a parallel strategy is taking shape. The prices are also quite high. All three models are priced below comparable offers from Amazon and Google.
If these models work well, the MAI family could quietly become the backbone of Microsoft’s entire AI product portfolio.
