From the source
Microsoft AI has introduced MAI Realtime 1, its first speech-to-speech model, aimed at enterprise voice agents and priced from $0.01 per audio minute.
- The company says the model outperforms GPT Realtime 2.1 on quality, speed and price across public benchmarks and
Microsoft product evaluations, placing it on the performance-price Pareto frontier for Big Bench Audio speech reasoning and the Tau-Voice agentic task completion benchmark.
- It handles interruptions, quick acknowledgments and pauses, taps a more powerful reasoning engine when a task needs deeper thinking, and recognizes requests over background noise or other speakers.
- It connects to enterprise systems and tools, adjusts tone and style through prompting, and is compatible with the
OpenAI Realtime API so developers can test it and migrate existing voice applications.
- Agoda and Reservations.ai are already testing it, and Microsoft is bringing it to Dynamics 365 Contact Center and to Dragon Copilot for AI-simulated patient conversations that help nurses practice clinical interactions.
- Developers can try it in MAI Labs or register for the Foundry limited-access preview, and preview customers won't be billed until public preview begins.
- Reasoning is billed separately at $1 per million input tokens and $5 per million output tokens.





