What does one minute of AI speech cost? Gemini, MAI, Voxtral and Sume
Convert per-token, per-character and per-minute TTS prices to a cost per finished minute, with a stated 900 characters per minute assumption.
At an assumed 900 characters per minute of speech, one finished minute costs about 1.4 cents on MAI-Voice-2.1-Flash, 2.0 cents on MAI-Voice-2.1, 1.4 cents on Voxtral TTS, 4.3 cents on Sume TTS, and 0.9 to 1.4 cents on Gemini 3.8 through December 31, 2026. The assumption is mine: 15 characters per second, so check your own scripts.
How the conversion works
Vendors price differently, so convert everything to dollars per minute of audio.
- Per character: price per 1,000 characters times 0.9.
- Gemini lists audio output per million tokens and gives about $0.00225 per 10 seconds for Flash TTS and $0.0015 for Flash-Lite through 2026, so multiply by 6.
- Text input tokens on Gemini add a small extra; the page lists $0.50 per million through 2026.
- Sume's job rounds up to a whole cent, with a one-cent minimum.
The table
All numbers are list prices, read from vendor pages and Sume's rate card.
| Service | List price | Per minute |
|---|---|---|
| MAI-Voice-2.1-Flash | $15 per 1M characters | $0.0135 |
| MAI-Voice-2.1 | $22 per 1M characters | $0.0198 |
| Voxtral TTS (API) | $0.016 per 1,000 characters | $0.0144 |
| Gemini 3.8 Flash-Lite TTS, to Dec 31 | $0.0015 per 10 seconds | $0.009 |
| Gemini 3.8 Flash TTS, to Dec 31 | $0.00225 per 10 seconds | $0.0135 |
| Gemini 3.8 Flash TTS, from Jan 1, 2027 | $0.0045 per 10 seconds | $0.027 |
| Sume TTS 1.0 | $0.0475 per 1,000 characters | $0.04275 |
What the price does not show
Sume's price covers a stored job record and a durable media URL. A raw per-character speech price excludes joining, captions and render.
A 60-second narration at 900 characters costs 5 cents on Sume after rounding, with a one-cent minimum per job and a 95-cent ceiling at 20,000 characters. Use dry_run on the MCP tool for an exact preview.
When the gap matters
At 1,000 minutes a month the difference between 1.4 and 4.3 cents is about $29. For a small studio that is noise; for an audiobook catalog it is a line item. Test voice quality first and price second.
Sources
Related posts
More in Pricing
- Unit-test a Seedance 2.5 cost estimator against six billed amounts
Your own cost estimate for Seedance 2.5 should match what Sume bills to the micro-dollar. A Python assert test with six known amounts from 480p to 1080p 30 s.
- What a 30-second Wan 3.0 job reserves on Sume: list price x 1.25
At submit Sume reserves provider list x 1.25. For Wan 3.0 at $0.05, $0.10, $0.20 per second, a 30-second job reserves $1.88, $3.75 or $7.50. Handle 402.
- Wan 3.0 bumper prices: 2 to 6 seconds at 480p, 720p, 1080p
A 2-second Wan 3.0 bumper costs $0.125 at 480p and $0.50 at 1080p on Sume. A 6-second one costs $0.375 to $1.50.
- Wan 3.0 draft at 480p, final at 1080p: no seed means a full re-render
Sume lists Wan 3.0 at $0.0625, $0.125 and $0.25 per second at 480p, 720p, 1080p. No v1 model takes a seed, so a draft is a prompt test, not a preview.
Written by Sume