MAI-Voice-2.1 at $22 per million characters: a 60-second script

A 60-second avatar script is about 900 characters, so MAI-Voice-2.1 at its listed $22 per million characters costs about $0.02 for the speech. Sume TTS 1.0 charges about $0.05 for the same text, and a 60-second Avatar 1.0 clip costs $11.04 or more, so the voice is under one percent of the bill either way.

What Microsoft lists

Microsoft published MAI-Voice-2.1 on 2026-10-01. Its model page lists two variants: MAI-Voice-2.1 at $22 per 1M characters and MAI-Voice-2.1-Flash at $15 per 1M characters. It covers 23 languages and is offered through the MAI Playground, Copilot Audio Expressions and Microsoft Foundry (Azure Speech).

Those are vendor list prices for the model. Sume does not list MAI-Voice in its TTS Router today. Sume TTS 1.0 uses Cartesia Sonic 3.6 and the TTS Router catalog is Cartesia Sonic only, so this post is a cost yardstick, not a Sume feature.

The arithmetic

A speaking rate of about 150 words a minute and roughly 6 characters per word, spaces included, gives about 900 characters for one minute. That is an assumption for planning, not a measurement. Sume TTS bills per transcript character, spaces and punctuation included, at the Cartesia list price of $0.000038 per character times the 1.25 house margin, which is $0.0000475 per character. Sume rounds a job up to the next cent.

Why the voice is not the budget line

Avatar 1.0 bills the rendered second, not the characters. At the standard tier a 60-second clip without a product image is 60 x $0.184 = $11.04. The plus tier is 60 x $0.245 = $14.70 and max is 60 x $0.55 = $33.00. Speech at $0.05 is 0.45 percent of the $11.04 standard clip. Saving a cent or two on the voice does not change which clip you can afford.

In an Avatar Video request you send the script and Sume speaks it with the avatar voice, so there is no separate TTS step to price. The speech cost matters when you generate audio first and then lip-sync a still to it, where you pay for the characters and then for the lip-sync seconds.

When a cheaper voice does matter

For a lip-sync route the audio must be hosted on Sume media before you submit, and Sume bills the audio seconds at the lip-sync rate. See the lip-sync price per second in the post linked below.

• Hundreds of long narrations with no avatar, where characters are the whole bill.
• Audio that you generate once and reuse across many clips.
• Language coverage: check the 23 listed languages against the ones you ship.
• Real-time agents, where the Flash latency tier is the reason to pick it, not the price.

What to do

Price the script, not the voice. Estimate seconds, multiply by the tier rate, and treat speech as a rounding error unless you narrate long audio without video. Check the Sume catalog for the live rate before a large batch.