Breaking

Microsoft Launches Three New MAI Voice Models for Real-Time Conversational Agents

October 2, 2026 at 12:00 EDT

Foundation Models
AI Agents

Microsoft unveiled three production-ready audio models on October 1 that enable developers to build fast, expressive voice agents with near-human conversational flow. The releases—MAI-Transcribe-2-Streaming, MAI-Voice-2.1, and MAI-Voice-2.1-Flash—address the full pipeline of real-time voice applications: listening, reasoning, and speaking.

October 2026 · Microsoft AI

Three New Voice Models for Real-Time Agents

Microsoft launches MAI-Transcribe-2-Streaming, MAI-Voice-2.1, and MAI-Voice-2.1-Flash for conversational agents with ultra-low latency and 60-language support.

First Transcript

100ms

Streaming latency

Languages

Transcription coverage

Voice Generation

150ms

Flash end-to-end latency

Speed Comparison: Voice Generation

Flash
150ms

Comparable
270ms

Turing Test Result

50.3% of listeners rated MAI-Voice equally or more human-like than human recordings

Availability & Pricing

Transcribe-2-Streaming:
\$0.54/audio hour

Voice-2.1:
\$22/1M characters

Voice-2.1-Flash:
\$15/1M characters

Platforms:
Foundry, Vercel, OpenRouter, Azure Voice Live

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.