News
MAI-Transcribe-2-Streaming: what Microsoft’s real-time speech model costs
No affiliate link on this page. We have no affiliate arrangement with Microsoft. Product links go to Microsoft’s official site.
Microsoft AI launched MAI-Transcribe-2-Streaming on October 1, 2026. It is a real-time speech-to-text model for voice agents, live captions, and dictation. If you build or buy software that listens to people talk, here is what Microsoft says it does, what it costs right now, and the one thing to check before you build on it.
What MAI-Transcribe-2-Streaming does
According to Microsoft’s launch post, the model streams transcripts in 60 languages and detects the spoken language automatically and continuously, so a caller can switch languages without a setting change.
Instead of waiting for a speaker to finish, it returns first guesses (Microsoft calls them “partials”) in just over 100 milliseconds, revises them as more audio arrives, and then commits a stable transcript. That matters for voice agents, which can start looking things up or calling tools mid-sentence.
Microsoft says the model ranks No. 1 for accuracy on both final and partial transcripts on Artificial Analysis, and that its internal tests show words appearing about twice as fast as its closest competitor for dictation and subtitles. Those are the vendor’s claims, so treat them as a reason to test, not as a result for your audio.
MAI-Transcribe-2-Streaming pricing
The launch price is $0.54 per hour of audio, which Microsoft describes as an introductory price through the end of the year. Two speech models launched alongside it:
- MAI-Voice-2.1, text-to-speech in 23 languages and 26 locales, where one voice keeps the same sound across languages, at $22 per 1 million characters
- MAI-Voice-2.1-Flash, a faster version for high-volume, low-latency work, at $15 per 1 million characters
Where you can use it
Microsoft lists Microsoft Foundry, the MAI Playground (which has a live voice-agent demo called Chatter), Vercel, and Azure Voice Live, with LiveKit marked as coming soon. The two voice models are also available through OpenRouter.
The downside: the price is only introductory
The $0.54 rate runs only through the end of the year, and the announcement does not say what the price will be after that. If you are pricing a product or a support budget on it, you cannot lock in your 2027 cost yet. We covered the same pattern in when a free preview becomes a bill.
Who it fits
- Teams building phone or chat voice agents that need to act before the caller stops talking
- Apps that show live captions or dictation for users who speak different languages
If you only need transcripts of a few recorded meetings a month, a developer model is more setup than you need. An app with transcription built in will be simpler. You can compare options on our tools page.
A quick check before you build
- Test it with your own recordings: your accents, your background noise, and the languages your customers actually use.
- Measure delay in your full setup, since your network and the voice reply add time on top of the model.
- Price your plan twice: once at $0.54 per hour and once with room for a higher rate after the introductory period.
FAQ
How much does MAI-Transcribe-2-Streaming cost?
$0.54 per hour of audio as an introductory price through the end of the year. Microsoft’s announcement does not list the price after that.
How many languages does MAI-Transcribe-2-Streaming support?
60 languages, with automatic and continuous language detection.
Where can I try it?
Microsoft points to the MAI Playground and Microsoft Foundry, along with Vercel and Azure Voice Live.
Sources
- Microsoft AI: Our first streaming transcription model debuts at no. 1 on Artificial Analysis (October 1, 2026; checked October 6, 2026)