How much is a 90-second AI voiceover: MAI-Voice-2.1 vs Sume TTS

A 90-second voiceover is roughly 1,350 characters, which costs about $0.0203 on MAI-Voice-2.1-Flash, $0.0297 on MAI-Voice-2.1 Standard and 7 cents billed on Sume TTS. The spread is five cents, so for a finished 90-second video the voice is the smallest line on the receipt.

The 1,350 figure is an assumption, not a measurement: 150 spoken words a minute, about 6 characters a word including the space, is 15 characters a second. Your script may sit at 1,100 or 1,600, so the table below shows all three.

The prices behind the table

Microsoft's MAI-Voice-2.1 model page lists $22 per 1M characters for the Standard model and $15 per 1M characters for Flash (read 2026-10-05). Sume's TTS 1.0 and TTS Router are character-metered at $0.0475 per 1,000 characters, which is Cartesia list price times 1.25, and each job is rounded up to a whole cent. Spaces and punctuation count on Sume.

The Microsoft columns are raw character math with no rounding, because the page does not say how a request is rounded. The Sume column shows both the raw figure and the billed cents.

What else a 90-second video costs on Sume

The voice is rarely alone. Sume's Timeline 1.0 renders at $0.10 per ceil output minute, so a 90-second output is billed as 2 minutes, or $0.20. A standalone caption job is $0.20 for videos up to 60 seconds under the current fixed estimate, and the caption page tells you to confirm the live price in the catalog for anything longer. A music bed from Music 1.0 is a fixed $0.125.

That makes a voice, a 2-minute-billed timeline render and a bed come to 7 + 20 + 12.5 cents on the 1,350-character script, before captions and any footage. Switching the voice to Flash would save about five cents of it.

Measure your own characters per second

Do not trust the 15 characters a second. Run the script once with timestamps.words: true on a Sume TTS job, read the last word's end time, and divide your character count by it. A slow, pause-heavy brand read can land well under 15, and a fast promo read well over.

Once you have your own ratio, you can predict the next script's cost and length from its character count alone. Sume's speed setting accepts a multiplier from 0.6 to 1.5, so the rate is adjustable, but the character count you pay for does not change with speed.

When the price gap matters

The point is to size the decision by volume, not by headline price per million characters.

• One 90-second ad: the gap is under 5 cents either way. Pick on voice quality and language coverage.
• A 52-week series: 52 reads at 1,350 characters is about $1.05 on Flash and $3.64 on Sume. Still small beside render and caption costs.
• A catalog of 10,000 reads at 1,350 characters: about $203 on Flash against roughly $700 on Sume at 7 cents each. Here the gap is real and worth testing.

Check before you commit

Take your real script, count its characters the way JavaScript counts string length (emoji and Hangul count differently from bytes), and compute the estimate. Compare the Sume cents with the Microsoft math, then listen to both reads. A 3-cent saving on a read that needs a retake is not a saving.

A worked receipt

Here is the 1,350-character case end to end on Sume: voice 7 cents, a 2-minute-billed Timeline render 20 cents, one Music 1.0 bed 12.5 cents. That is 39.5 cents before captions, which add $0.20 for a clip up to 60 seconds, so check the catalog for a 90-second output. Three voice takes while you choose a read add 14 cents, which is less than the render.

The practical rule is to audition freely and render once. Voice takes cost cents, so run several, listen, and only then commit the timeline minute and the caption job.