MAI-Voice 2.1 by Microsoft: the AI voice half of listeners find human
MAI-Voice 2.1 by Microsoft: an AI voice 50.3% of listeners rated as human as a person. Languages, pricing and limits.
MAI-Voice 2.1 is Microsoft AI's new synthetic voice model, released on October 1, 2026. According to Microsoft, 50.3% of listeners rated its voice equal to or more human than a recording of a real person. Here is what it does, what it costs and what it means.
What MAI-Voice 2.1 is and what the test says
Microsoft reports that half of listeners judged the voice as human as, or more human than, a real recording. That figure comes from the company itself, so read it as such: it is not an independent audit. Still, it leaves a practical rule: hearing a human-sounding voice no longer proves a person is speaking.

What it can do: three key capabilities
- Speak 23 languages, at 22 dollars per million characters.
- Clone a voice from just 5 seconds of audio, with no studio or voice actor.
- Flash version: responds in 150 milliseconds and costs 15 dollars per million characters.
Beyond generating speech, the model also transcribes live in 60 languages.
Launch pricing: watch the date
Microsoft offers a price of 54 cents per hour, but it is launch pricing only, valid until the end of the year. If you are budgeting a project, don't treat it as the permanent rate.

How to apply it in your business?
If you quote voiceovers for clients (ads, phone support, courses or videos), this kind of synthetic voice changes the conversation: the cost is no longer the voice, it is the script, meaning what it says. Compare the per-million-character rates, decide whether the Flash version's speed is enough, and confirm the current price before closing a budget. And if you clone a voice, make sure you have permission from the person who lends it.
Microsoft AI also moved transcription pricing: read the full breakdown of Microsoft AI cutting transcription to 10 cents per hour.
Frequently asked questions
What is MAI-Voice 2.1?
It is Microsoft AI's synthetic voice model, released on October 1, 2026.
How many languages does it speak?
It speaks 23 languages and transcribes live in 60.
How much does it cost?
The standard voice costs 22 dollars per million characters and Flash costs 15. The 54-cents-per-hour price is launch pricing, until the end of the year.
How much audio does it need to clone a voice?
According to Microsoft, 5 seconds of audio.
Conclusion
MAI-Voice 2.1 shows that AI voices are now mistaken for human ones by half of listeners, according to Microsoft. Save this summary for the next time you quote a voice for a client.