# Voice One vs Voice Turbo — expressiveness against 2 credits > Voice One costs 3 credits per 100 characters and Voice Turbo costs 2 — what the extra credit buys, plus the arithmetic for a chapter and for a chat reply. Published 2026-08-31 · eroq.ai — canonical: https://eroq.ai/blog/voice-one-vs-voice-turbo The two eroq speech models differ by exactly one credit per hundred characters, and that one credit is the whole argument. [Voice One](/models/eroq-voice-one) is the expressive flagship at 3 credits per 100 characters. [Voice Turbo](/models/eroq-voice-turbo) is faster, at 2. Same roster, same clones, same MP3 coming out the other end. Which means the decision is never about what a voice *can* sound like — it is about how much performance a given piece of text actually needs, and how many characters you are about to render. ## What is identical Start with the shared half, because it is most of it. - **The same voices.** Aria — warm, intimate, female. Orion — low, calm, male, across 13 languages. Both models render both. - **The same clones.** A voice you cloned from your own audio works on Voice One and Voice Turbo without re-cloning. Details in [the voice cloning guide](/tools/ai-voice-cloning). - **The same controls.** Speed and expressiveness sliders, and punctuation-driven delivery, on both. - **The same output.** MP3. - **The same Characters.** A Character with a voice attached — roster or cloned — speaks on either model. So switching models is not a migration. It is a dropdown in [the Voice studio](/studio/voice) or one string in `audio/speech` on the API. Nothing else in your setup moves. ## What the extra credit buys Expressiveness, and the word is doing precise work. Voice One is the flagship for performance — the model to reach for when a line has to be *delivered* rather than read out. Turbo is the faster one, and speed in a speech model is not a vanity metric: in anything interactive, the gap between asking and hearing is the product. Put practically: **Voice One earns its credit when the text is performed.** Narration a listener will spend twenty minutes with. A character's first line. A trailer. Anything where a flat read would be noticed as a flaw rather than tolerated as a convenience. **Voice Turbo earns its credit back when the text is functional.** Notifications, menu prompts, UI readouts, bulk drafts, long documents nobody is going to savor, and — crucially — chat replies, where waiting is worse than a slightly plainer read. There is a third case that people underrate: **drafting**. Render your script on Turbo while you are still cutting words, then re-render the final pass on Voice One. You pay 2 credits per 100 characters for every version you throw away and 3 only for the one you keep. ## The arithmetic, done twice Credits run about a cent each at the entry pack, so the math below converts easily. Both models are priced per 100 characters of input text. ### A chapter Take an audiobook chapter of 3,000 words — call it 17,000 characters with spaces. That is 170 blocks of 100. - **Voice One**: 170 × 3 = **510 credits**, roughly $5.10. - **Voice Turbo**: 170 × 2 = **340 credits**, roughly $3.40. - **Difference**: 170 credits a chapter. Across a twelve-chapter book, 6,120 credits versus 4,080 — about $20 apart for the whole title. For a finished audiobook, that is not a decision, it is a rounding error. Render the book on Voice One. Where Turbo earns its place in this job is the six drafts you made while still rewriting chapter three. More on the format side in [audiobooks and podcasts](/use-cases/audiobooks-podcasts). ### A chat reply Now the other extreme. A voiced reply in a companion or roleplay product runs maybe 220 characters — 2.2 blocks. - **Voice One**: about 6.6 credits a reply. - **Voice Turbo**: about 4.4 credits a reply. Two credits sounds like nothing until you multiply. At 500 voiced replies a day, that is roughly 3,300 credits a day on Voice One against 2,200 on Turbo — about 99,000 versus 66,000 credits a month. Same voice, same clone, 33,000 credits of difference, plus the latency advantage on the model that also happens to be cheaper. This is the single clearest split in the eroq voice roster: **long-form goes to Voice One, high-volume interactive goes to Voice Turbo**, and the reason is the same in both cases — you are paying per character, so the model choice compounds exactly as fast as your character count does. The broader version of that argument is in [giving AI characters a voice](/blog/tts-for-ai-characters). ## The script is the setting Both models read punctuation as direction, which means most of the control you have lives in the text rather than in a slider. Write the script the way you want it performed: > She stopped at the top of the stairs. Not because she'd heard anything — because she hadn't. > > "Hello?" > > Nothing. The house did that thing old houses do, where the silence has a shape to it. > > "I know you're down there." > > And then, quieter, almost to herself… "I know you're down there." Every device in that block is doing a job. The em dash holds a beat. The short paragraph isolates the line so it does not get swallowed. The ellipsis slows the last delivery. The repetition, set apart, comes out differently the second time because the punctuation around it changed. Then the sliders, sparingly. **Speed** for pacing — a shade under default for narration, default for dialogue. **Expressiveness** for how far the read leans into the punctuation you wrote. Pushing expressiveness up on a flat script does not make it dramatic; it makes it strange. Fix the script first. A test worth doing before you commit a project: take your worst paragraph — the one with a phone number, an acronym and a brand nobody can pronounce — and render it on both models with the same voice. That comparison costs a few credits and tells you more than any demo reel. The API side is documented at [/docs/speech](/docs/speech), and [text-to-speech](/glossary/text-to-speech) has the term itself if you need to hand this to someone. ## Pick in one line - Long-form, performed, final pass → **Voice One**. - Interactive, high volume, or a draft you will re-render → **Voice Turbo**. - Unsure → render the same 200 characters on both and listen. It costs about 11 credits to settle it. ## FAQ ### How much do Voice One and Voice Turbo cost? Voice One is 3 credits per 100 characters, Voice Turbo is 2 credits per 100 characters. Both output MP3 and both work with the full voice roster and with your cloned voices. ### Do cloned voices work on both models? Yes. A voice cloned from your own audio is available on Voice One and Voice Turbo without re-cloning, and Characters keep whichever voice they were given on either model. ### Which model should I use for a chat app? Voice Turbo in almost every case. It is the faster model and it costs a third less per character, and in an interactive product latency is more noticeable than the last measure of expressiveness. Render the same line on both and decide with your ears — [open the Voice studio](/studio/voice).