Microsoft has launched its first artificial intelligence (AI) models that it has fully built by itself. On Thursday, the company introduced two new foundation models made to work with its Copilot services. This is the first time Microsoft has created large AI models from the beginning, instead of only depending on OpenAI.
The first model is called MAI-Voice-1. It is a voice AI system that can create speech that sounds natural and full of expression. Unlike normal text-to-speech tools that just read text in a flat way, this model can actually talk like a real person. It changes its tone, pitch, and speed automatically depending on the situation. Microsoft says it is very fast and can make one full minute of audio in less than a second on just one GPU.
Microsoft has already started using MAI-Voice-1 in a new feature called Copilot Daily. In this, an AI host reads out the top news stories of the day like a podcast and also explains them in simple conversation. This model is also available in Copilot Labs experiments, where users can type any text, select a voice or style, and the AI will read it aloud in that way.
The second model is called MAI-1-preview. It is a text-based AI that can understand instructions and answer everyday questions. It is a general AI system that gives replies in normal, human-like text. Microsoft trained it using about 15,000 Nvidia H100 GPUs with a “mixture-of-experts” design, which makes it more efficient at handling tasks.
Right now, MAI-1-preview is still being tested. Users can try it through Microsoft’s crowdsourced testing platform called LMArena, and some selected users can access it through an API. Microsoft plans to add this model into Copilot soon, so it can be used for text-based tasks across apps.
This launch marks a big change in Microsoft’s AI plans. Since 2023, the company mostly used OpenAI’s GPT models inside Copilot, along with some smaller custom models. But until now, it never made its own large foundation models. With these new models, Microsoft is showing that it wants to create more AI on its own, while still working with partners like OpenAI.
For users, this means Microsoft Copilot will become more powerful and feel more natural. With MAI-Voice-1, AI voices will sound smoother and more human-like instead of robotic. With MAI-1-preview, Copilot will be better at answering questions, following instructions, and helping with tasks in apps like Word, Excel, Outlook, and Teams.
Experts say this move will also help Microsoft compete with other big AI companies such as Google DeepMind, Anthropic, Meta, and xAI. By building its own foundation models, Microsoft will have more control over speed, cost, and future improvements. It also gives the company a backup if it wants to depend less on OpenAI in the future.
Right now, MAI-Voice-1 can be used in Copilot Daily and Copilot Labs, while MAI-1-preview will be released step by step in the coming weeks. With this step, Microsoft is preparing for more advanced AI tools in its products, so users can expect Copilot to sound smarter and act more helpful soon.