Limits
Three independent layers keep usage predictable: hard limits on each request, tokens on your account, and a request rate on each key.
Per request
| What | Limit |
|---|---|
speak text | 5 000 characters (1 200 for Chinese, Japanese and Korean) |
podcast text | 100 000 characters (about 80 minutes of audio) |
| Reference recording for a clone | WAV or MP3, up to 10 MB — three to ten seconds of clean speech is enough |
| File to transcribe | up to 5 GB and eight hours; WAV, MP3, M4A, MP4, OGG, WebM, FLAC, AAC |
project | letters, digits, spaces and . _ -, at most 60 characters |
| List pages | 20 items by default, limit up to 100, then cursor |
Line breaks and runs of spaces collapse to a single space before characters are counted, and markdown symbols, URLs and emoji are removed. Pasted text costs no more than it reads.
Per account: tokens
Speech, streaming, transcription and voice clones are paid in tokens — free tokens every day plus any balance on your account. See Tokens for the rates and when they are charged. A request you cannot afford answers 402 with code INSUFFICIENT_TOKENS; a clip that fails gives its tokens back.
Per key
API keys allow 1 request per second (bursts of 3) and 1 000 calls per day. Going over answers 429 {"message":"Too Many Requests"} from the gateway. Polling a clip counts as a call, so poll every second or two, not in a tight loop.
Links
- Signed audio URLs (
url,sampleUrl, transcript files) last one hour. Fetch the object again for a new one. - Upload URLs for a reference recording or a file to transcribe last 15 minutes.
- Share links (
shareUrl) never expire; revoke them withDELETE /clips/{clipId}/share.