Microsoft has introduced its first in-house (internally developed) artificial intelligence models: MAI-Voice-1 and MAI-1-preview. The release marks a strategic step in the company’s effort to strengthen its position in the global AI landscape. Mustafa Suleiman, head of Microsoft AI, explained the importance of maintaining independent expertise in developing advanced AI systems.
Advertisеment
MAI-Voice-1
MAI-Voice-1 is a speech-focused model designed for high efficiency. It can generate one minute of audio in under a second using a single graphics processing unit. Microsoft already uses this model in consumer-facing features such as Copilot Daily, where an AI presenter delivers news summaries, and in generating conversational podcast-style content. Users can interact with MAI-Voice-1 through Copilot Labs, where text input is converted into spoken output.
MAI-1-preview
The second model, MAI-1-preview, is a text-based system trained on approximately 15,000 Nvidia H100 GPUs. In benchmark tests on the LMArena platform, it ranked 13th, placing it behind models such as DeepSeek R1, Claude Opus 4, GPT-4.1, and Grok 3 Preview. Microsoft noted that MAI-1-preview may serve as the foundation for future versions of Copilot, the AI assistant integrated into Windows and Office applications.
The company highlights that MAI-1-preview is optimized for consumer applications, with capabilities in following instructions and responding to everyday queries. Current versions of Copilot rely on OpenAI’s large language models, but Microsoft plans to incorporate its own technology into the assistant over time.
The future
Microsoft AI is focusing on creating models tailored to specific user needs. The team is working on a range of specialized systems designed to improve performance across different use cases. Testing of MAI-1-preview is now underway on public benchmarking platforms.
Suleiman has previously stated that Microsoft’s AI development prioritizes consumer utility over enterprise deployment. He cited access to extensive consumer data in advertising and telemetry as a key advantage in building practical AI solutions. Just recently, he introduced the concept of Seemingly Conscious AI (SCAI), warning that AI systems simulating human-like awareness may lead to psychological and societal challenges, including blurred perceptions of reality and misplaced calls for AI personhood.
Support us
Winaero greatly relies on your support. You can help the site keep bringing you interesting and useful content and software by using these options:
