Microsoft’s AI branch unveiled its initial proprietary AI models on Thursday: MAI-Voice-1 AI and MAI-1-preview. According to Microsoft, the MAI-Voice-1 model is capable of producing a full minute of audio in less than a second using only a single GPU. Meanwhile, MAI-1-preview is described as a sneak peek into future features that will be integrated into Copilot.

Users curious about MAI-Voice-1 can experiment with it on Copilot Labs. This platform allows you to input desired phrases and modify the voice characteristics and tone of the speech. Alongside this, Microsoft also presented MAI-1-preview, claiming it was developed using approximately 15,000 Nvidia H100 GPUs. It’s designed for individuals needing an AI model that can understand directions and provide useful feedback to common inquiries.

Mustafa Suleyman, head of Microsoft’s AI unit, discussed in an episode of Decoder last year that their internal AI models are not primarily directed at enterprise applications. “The rationale is to develop high-performance solutions for consumers, optimizing around our own use cases,” Suleyman stated. “With extensive data on advertising and consumer behavior, my aim is to create models that truly resonate as consumer companions.”

The company intends to deploy the MAI-1-preview model for various text applications within its Copilot AI assistant, which presently utilizes OpenAI’s extensive language models. Furthermore, Microsoft has initiated public trials of the MAI-1-preview model on the AI benchmarking platform LMArena.

“We are aiming high for our future endeavors,” stated Microsoft AI in its blog post. “Our focus is not only on continued progression but also in coordinating diverse specialized models to meet varied user requests, which we believe will bring substantial benefits.”