MAI-Voice-2-Flash is a low-latency speech model designed for contact centres and voice agents that cannot afford a pause. It is Microsoft's speed tier of the MAI voice line, announced by Mustafa Suleyman on 23 July 2026: the heavier MAI-Voice-2 still handles long-form narration and speaker consistency, while Flash targets turn-taking and throughput.
Core Features
- Roughly twice as fast as MAI-Voice-2 while holding the same prosody and acoustic quality, according to Microsoft.
- Powers Dynamics 365 Contact Center, where Microsoft reports GPU costs falling by as much as 89 percent.
- Wired into Azure Voice Live for speech-to-speech agents, so you skip stitching separate listen, reason, and speak services together.
- Covers 15 languages across 18 locales in Microsoft documentation.
- Voice cloning is available but gated, and only from a consented reference clip of 5 to 60 seconds.
Use Cases
- IVR and call deflection, where a two-second pause is the difference between a resolved ticket and a hang-up.
- High-volume support desks measured in millions of conversations a quarter.
- Booking assistants and workplace agents that need an immediate spoken reply.
Pricing
$15 per million characters in Microsoft Foundry public preview, which Microsoft puts at 32 percent below MAI-Voice-2. Preview is not general availability, so confirm your region and access status before you commit to a launch date. Enterprise consumption bills through Azure alongside the rest of the MAI catalogue.
Our Take
Best for teams already on Dynamics 365 or Azure who are optimising cost per call rather than chasing the most expressive read. The trade-off is range: ElevenLabs and the full MAI-Voice-2 still win on long narration, and 15 languages looks thin next to rivals covering thirty or more. Compare speech tools at aifreetool.site/tool-category/ai-audio-tools/.
