Which text-to-speech model, and what each one charges by

LUVI's catalog carries eight text-to-speech models, and the first thing to understand about them is not the voices—it is the meter. Most of them charge per thousand characters, linearly, with no minimum and no rounding to a block. On September 22, 2026 the published rates ran from 30 credits per thousand characters up to 300, which means the same script can cost ten times as much depending only on which model reads it.

What do they actually cost?

Per thousand characters, on September 22, 2026:

• Grok Speech (tts-v1): 30 credits
• Gemini 2.5 Flash TTS: 80 credits
• MiniMax Speech 2.6 Turbo: 96 credits
• Gemini 2.5 Pro TTS: 160 credits
• MiniMax Speech 2.6 HD: 160 credits
• ElevenLabs v3: 200 credits
• Gemini 3.1 Flash TTS: 300 credits

To make that concrete, we quoted one 311-character paragraph on two of them. Grok Speech came to 10 credits; ElevenLabs v3 came to 63. The arithmetic is exactly linear—rate times characters, divided by a thousand, rounded up to a whole credit—so a short line is genuinely cheap on every model in the list, and a long script is where the rate starts to matter.

Note that the newest Gemini tier is also the most expensive of the seven listed above. Newer is not cheaper here.

Which one is billed differently?

Seed Audio 1.0, and the difference is worth knowing before you look at its estimate. It is billed by the minute rather than by the character, and the number you see before you run is a two-minute hold: our quote came back at 572 credits no matter how short the text was.

That is not what a short line costs. The charge settles at the real duration afterwards, the way every generation on LUVI settles at the provider's real bill—so a twelve-second read will not take 572 credits off your balance. The help center explains why the charge can differ from the estimate in general.

What does the extra rate buy?

Control, mostly, and it is worth checking the parameter list before assuming.

• ElevenLabs v3 exposes twenty-one voices, a stability dial from 0 to 1, an ISO language code to force a language, and a text-normalization switch that decides whether numbers get spelled out. Four of its voices are described as multilingual; the others are tuned to their own language and will cascade-update the language field when you pick them.
• The Gemini tiers are the same family at three price points, so the tier is the decision rather than the dialect.
• Grok Speech is the cheapest by a wide margin, which makes it the natural choice for drafts, temp tracks and anything you will replace later.

There is no single "best voice" answer here, because the thing you are judging is a performance. The practical method is to run the same two sentences through two or three of them at a few credits each and listen.

What about the other direction?

Speech to text is on the same page and priced differently again: Grok's stt-v1 is billed per minute of audio, at 4 credits a minute. Transcripts come back as text results in your library rather than as audio files, and they are deliberately excluded from the reference pickers—a transcript is never model input.

How we checked

• Models: Grok Speech tts-v1, ElevenLabs v3, Seed Audio 1.0, with the Gemini TTS, MiniMax Speech and Grok stt-v1 rates read from the catalog.
• Settings: one 311-character paragraph, at each model's default voice and settings.
• Date and what was read: September 22, 2026, re-checked October 20, 2026, with the same results—live credit quotes and the models' parameter contracts in LUVI.
• Results: 10 credits on Grok Speech and 63 on ElevenLabs v3 for the same paragraph; 572 credits quoted on Seed Audio 1.0 regardless of text length, which is a two-minute hold rather than a price. Published rates per thousand characters ran 30, 80, 96, 160, 160, 200 and 300.
• Drawbacks: we compared meters and parameters, not voices. Nothing here says which model sounds better, and no audio was generated for this post.

In LUVI

Open the models page and filter to audio, then write your two test sentences once and run them through the two cheapest models before you commit a whole script to one voice. The credit estimate above the Generate button updates as you type on this family—it is one of the few places in the catalog where the text itself moves the price, which is exactly what you want when the text is the product.

New accounts start with 200 credits, and verifying a phone and a card adds up to 1,500 more. You can create a LUVI account and audition several voices on your own script.