Skip to main content
Breeze uses credits shared across Studio and the Developer API. Text to speech is metered by text length. Voice Design and Voice Clone each cost 100 credits per generated preview on both Web and the Developer API.

​
How we meter

  • Text to speech: per character of the request text. Chinese, Japanese and Korean characters (Han ideographs, kana and Hangul) count as two billable units each; every other character (Latin letters, digits, spaces, punctuation including CJK punctuation, emoji) counts as one. 10 billable units = 1 credit.
  • Realtime text to speech: the same per-character rule applied to each appended turn’s text, committed when the turn settles; turns that never produce audio are not billed. See the realtime guide.
  • Voice design: 100 credits per successfully generated preview, regardless of preview text length or language. When preview_count is omitted, Breeze generates one preview. A request for 3 previews reserves 300 credits; if only 2 succeed, 200 credits are charged and 100 are released. Failed requests release their reservation.
  • Voice cloning: 100 credits per clone generation, regardless of sample or preview text length. Web reference-audio analysis costs an additional 1 credit per analysis. Saving an existing generated preview does not incur another generation charge.
  • Voice localization: 100 credits per localization generation, regardless of target language or preview script length. Saving the resulting preview does not incur another generation charge.
  • List, get, and delete catalog reads: not billed.
For Voice Design and Voice Clone, billable_units is the number of generated previews. Text character counts in usage metadata are diagnostic and do not affect these charges. Text-to-speech example: 你好世界hello is 9 characters, 4 of them CJK, so it bills 4 × 2 + 5 × 1 = 13 billable units = 1.3 credits. In usage records, text_characters is the raw character count (9) and billable_units is the charged unit count (13).

​
Generation concurrency

Each plan includes a maximum number of simultaneous generations. A generation is admitted or running synthesis work. Studio and Developer API jobs share the same per-user pool. Requests over the limit fail fast with 429 GENERATION_CONCURRENCY_EXCEEDED and Retry-After. Sync text to speech, streaming text to speech, async text to speech, voice design, voice cloning, and voice localization all count toward these limits. Voice design counts each requested preview as one generation.

​
Where to see your spend

Balance, last 30 days of usage, and per-request charges are on the usage page. For teams or shared accounts, assign a hard credit limit to each API key on the API keys page. Key budgets cap how much of the shared account balance one employee, application, or environment can consume. They do not preallocate credits, and resetting a key budget does not refund the account balance. See Authentication for the exact semantics.

​
Free credits and top-ups

Credits are shared across Studio and the Developer API. Top up on pricing.

​
Plan your usage

Rate limits

Understand concurrent generation limits and retry guidance.

Usage page

Review balance, recent usage, request charges, and credit consumption.

Get usage metering

Query aggregated usage with daily buckets and recent requests programmatically.

Get account balance

Check the current credit balance and recent top-ups before large batch runs.

Text to speech guide

Choose sync, streaming, or async delivery with cost and concurrency in mind.

Errors

Handle insufficient credits, concurrency limits, capacity, and retryable failures.