How billing counts what you use
Every billable action, its unit, what is free, and the cases people ask about: repeats, library lines, cache hits, failures and preparation.
Billing is live: your free allowance applies first, then you pay as you go. This page describes how usage is counted so you can plan. List prices may change; we will say so here.
The units
| Product | Counted in | List price | Free per account |
|---|---|---|---|
| Catalog voices (library and live) | Characters of the text you send | $0.70 per 1M | 100,000 characters (shared with your own voice) |
| Your own voice | Characters | $6 per 1M | same 100,000 |
| Styled voice (private preview, off in production) | Characters | $17 per 1M | same 100,000 |
| Speech to text | Seconds of audio, listed per minute | $0.0015 per minute | 60 minutes |
| Voice agents (beta) | Turns (one question and one answer) | $10 per 1,000 turns | 200 turns |
The cases people ask about
| Situation | What is counted |
|---|---|
| A library phrase through the API | Its length in characters, every time you request it |
| The same sentence again | Billed again. A repeat is faster, not free |
| A brand-new sentence generated live | Its length in characters, once per request |
| Your own voice, first request | Characters, billed; the audio is generated and returned, not yet stored |
| Your own voice, second request for the same text | Characters, billed; the audio is stored on the second request and later requests reuse it |
| A request that fails (for example upstream unavailable) | The request is recorded; no characters are charged |
| POST /v1/prepare | Not billed. Speaking the prepared sentences later is billed as usual |
| Audio you stored yourself and replay without calling the API | Nothing |
| Speech to text partial results (interim) | Billed as audio seconds because the buffer is re-transcribed |
| Searching the library (/v1/search) and listing voices | Not metered as characters |
Free allowance behaviour
Accounts on the free plan can use up to 100,000 characters in total; after that the API answers 402 free_allowance_used until a paid plan is active. Paid accounts are not capped by the allowance. Speech to text and agent turns have their own allowances (60 minutes, 200 turns).
What changes with the price
- Cheap characters, so repeats are fine. At $0.70 per million, a 40-character line costs about $0.000028, so repeating lines through the API is rarely the problem; heavy replay is better served from your own copy (guide).
- Your own voice is about 8.6 times the catalog rate ($6 against $0.70). Use catalog voices for bulk notifications and your own voice for the lines where the voice itself matters.
- Transcription is billed on audio length, not on words, so silence costs the same as speech.
Seeing your usage
The portal shows a 30-day usage chart, your plan and a free-allowance bar. Webhooks can tell you about key events. A daily usage report goes to Stripe once billing is live.
Compare with other providers in their own units: text to speech · speech to text · pricing and calculator.