Newshot

Microsoft's first speech-to-speech model: MAI Realtime 1 debuts, starting at $0.01 per minute

AI summary

At Ignite 2026 on November 17, Microsoft announced MAI Realtime 1, its first speech-to-speech model, positioned as real-time voice AI for real enterprise scenarios. It handles natural conversation details such as interruptions, quick confirmations, and waiting for the other person to finish, and can understand requests in noisy environments or when multiple people speak at once. Base speech-to-speech usage is priced at $0.01 per audio minute, with reasoning billed separately at $1 per million input tokens and $5 per million output tokens.

Translation(model translation, not the source's own text)

Microsoft's First Speech-to-Speech Model: MAI Realtime 1 Debuts, Starting at $0.01 per Minute

IT Home, November 18: At the Ignite 2026 conference held yesterday (November 17), Microsoft announced its first speech-to-speech model, MAI Realtime 1, positioned as real-time voice AI for real enterprise scenarios. The model can handle natural conversation details such as interruptions, quick confirmations, and waiting for the other party to finish speaking, and can understand user requests in noisy environments or when multiple people are speaking at once.

On pricing, Microsoft says the base speech-to-speech usage fee for the model is $0.01 per audio minute (IT Home note: about 0.067 yuan at current exchange rates); if reasoning is involved, it is billed separately at $1 per million input tokens (about 6.7 yuan at current exchange rates) and $5 per million output tokens (about 33.4 yuan at current exchange rates).

Microsoft says MAI Realtime 1 can not only respond to speech in real time, but also call on a stronger reasoning engine when a task requires deeper thinking, and connect to enterprise systems and tools to complete complex operations such as bookings and customer service handling; to adapt to different customers and industries, developers can also adjust its tone, style, and behavior through prompts.

Microsoft says that in public benchmarks and internal product evaluations, MAI Realtime 1 outperforms GPT Realtime 2.1 on quality, speed, and price, and sits on the "performance–price" frontier on the Big Bench Audio speech reasoning and Tau-Voice agentic task completion benchmarks.

Test data released by Microsoft shows that MAI Realtime 1 achieves 98.9% accuracy on Big Bench Audio, with an input audio cost of about $1.35 per hour (about 9 yuan at current exchange rates); by comparison, GPT-Realtime-2 High is 96.6% at $4.14 per hour (about 27.7 yuan at current exchange rates), and Grok Voice 2.0 is 97.2% at $4.80 per hour (about 32.1 yuan at current exchange rates).

The model is already compatible with the OpenAI Realtime API, making it easier for developers to test and migrate existing voice applications. Companies such as Agoda and Reservations.ai have begun trials.

Microsoft also plans to use it in Dynamics 365 Contact Center customer service scenarios, as well as for AI-simulated patient conversations in Dragon Copilot to help nurses practice clinical communication.

Related articles

  • Why are some Microsoft products always a bit short on polish? Azure engineering lead says large customer needs come first
  • No subsidies, no shifting costs onto others: Microsoft promises to pay for its AI expansion's electricity bills itself
  • Microsoft officially releases .NET 11: less time spent on development, more complete AI tools
  • AI boom collides with resource disputes as Microsoft's under-construction data center in the Netherlands faces protests
  • Borrowing money to do AI: Microsoft reportedly plans to raise at least $25 billion through a US bond issuance
  • "The world's second-best image editing AI model": Microsoft MAI-Image-2.7-Preview debuts, beating Google, Meta, xAI

The translation is machine-generated and individual terms may be off. Nothing is rewritten or cut; the original is authoritative.

Published by IT之家
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗