Microsoft is reducing its reliance on OpenAI by introducing two of its own artificial intelligence models: MAI-Voice-1 and MAI-1-Preview. The new models for Copilot hint that in-house development is a growing focus for the productivity leader. They help keep pace with competitors in the fast-moving AI market.
What’s Happening & Why This Matters
For years, Microsoft has leaned on its multibillion-dollar partnership with OpenAI to power Copilot. This is the assistant integrated across Windows, Office, and its cloud tools. Now, the company has unveiled MAI-Voice-1, a text-to-audio generator. It also introduced MAI-1-Preview, a large language model fully developed in-house. Copilot’s in-house models designed to enhance functionality.
MAI-Voice-1 is built to create audio clips at a rapid speed. According to Microsoft, the model can generate clips in less than a second on a single GPU. It already powers Copilot Daily, an audio news and tips feed. It also powers Copilot Podcasts, which simulate conversations between two AI voices. In testing, audio generation took approximately three to four seconds, instead of under one second. Nevertheless, users were impressed with the quick response. The model also offers mood customisation, such as “”Emotiv”” for dynamic tones and “”Stor”” for book-style narration.
The second release, MAI-1-preview, is Microsoft’s first fully in-house trained large language model. It was pre- and post-trained across 15,000 Nvidia H100 GPUs and integrates multiple smaller LLMs. Introducing Copilot’s in-house models, the company describes it as a system designed to handle instruction-following and practical queries for consumers. The model is currently being tested on LMArena, a public benchmarking platform, before an extended roll-out for text-based use cases.
Microsoft says the MAI team has been working on these models since 2024. They frame them as the foundation for “applied AI as a platform for category-defining and deeply trusted products that understand each of our unique needs.” This effort also plays into the recruiting strategy. Microsoft used the announcement to encourage AI talent to join its MAI division.
Insiders view the decision as an important step. It helps Microsoft gain more autonomy in AI development. With competitors like Google, Anthropic, and xAI advancing rapidly, the company’s ability to produce in-house models gives it more control over future innovation. Copilot’s in-house models set Microsoft more favorably in AI markets.
TF Summary: What’s Next
The launch of MAI-Voice-1 and MAI-1-preview clears Microsoft’s intent to move beyond its OpenAI dependency. They expand their AI footprint with internally developed tools. The text-to-audio model adds versatility to Copilot. Meanwhile, the in-house LLM promises more specialised and reliable responses for everyday users.
If testing goes well, expect these models to develop & deploy more Copilot functionality in the coming months. In-house development boosts Microsoft’s flexibility in how its AI interacts with consumers and businesses.
— Text-to-Speech (TTS) provided by gspeech