Skip to main content
Bland Speech generation is billed per character against your wallet balance. Limits are per organization, and you can request a raise for each of them.

Default limits

Topping up your wallet does not change these limits. To go higher, see Raising your limits.

Instant voice clones

Every organization can create one instant clone by default, so you can try cloning right away. Instant clone limits are counted per organization against your current-generation (V3) voices. Voice clones created with earlier model generations don’t count against the limit. If your limit is 10, that’s 10 active V3 clones, and deleting one frees the slot. The limit applies when creating a new clone; existing voices are never affected by a limit change. Limits are raisable; see Raising your limits.

Concurrency responses

Your concurrency limit is the number of speech generations you may have in flight at the same moment, across every speech surface, including WSS /v2/tts/ws, POST /v2/tts, POST /v2/audio/speech, the /v1/speak family, and shared audio links (which count against the link owner’s organization). A successfully initialized /v2/tts/ws connection occupies one slot until the socket closes, including time between turns. Close idle conversation sockets promptly. The endpoint keeps its synthesis connection warm for low latency, so it reserves capacity for the full session. HTTP speech endpoints respond 429 with a Retry-After header when no slot is available. POST /v2/tts uses its standard error shape:
The /v1/speak family and shared audio links return:
/v2/tts/ws reports the same condition during init with a rate_limited error frame, then closes without sending ready. This is not ordinary rate limiting. A rate limit meters requests per unit of time; the concurrency limit counts reserved speech capacity. An HTTP slot frees when its request finishes. A /v2/tts/ws slot remains reserved until that socket closes. The correct client behavior is:
  • For HTTP, wait the number of seconds in the Retry-After header, then retry.
  • For WebSocket sessions, wait before opening a replacement connection.
  • If the cap remains full, wait for an HTTP request to finish or close an idle WebSocket before retrying.
Avoid tight retry loops. They cannot make a slot free faster, and the wait is typically only a few seconds.

Professional clones

Professional (studio-trained) voices follow a draft / live model:
  • You may hold unlimited drafts. Drafts can be trained and previewed in the studio, but cannot be downloaded, used through the API, or used in calls.
  • You may promote up to your live limit. The default live limit is 0, so request a raise before you promote your first professional voice. Only live voices are usable everywhere.
  • Training a professional voice requires passing Bland’s consent and rights review for the voice being cloned.

Raising your limits

For instant clones and live professional clones, use the Request a raise option in the studio (on the billing page when you’re at your limit, or in the clone dialog); requests are reviewed by hand and answered by email. For concurrency, email tts@bland.ai with your organization ID and expected volume, and those are sized case by case. Raises are applied directly to your organization, no plan change required.
Docs for agents: llms.txt