specification: API Commons Rate Limits
specificationVersion: '0.1'
schema: https://raw.githubusercontent.com/api-evangelist/interface-research/main/schema/api-commons.yml#/$defs/RateLimits
provider: Cartesia
providerId: cartesia-ai
created: '2026-07-02'
modified: '2026-07-02'
reconciled: true
tags:
- AI
- Voice AI
- Text to Speech
- Speech to Text
- WebSocket
- Rate Limiting
- Quotas
- Concurrency
description: >-
  Cartesia enforces concurrency-based limits rather than classic
  requests-per-minute quotas. Each plan caps the number of simultaneous
  TTS requests, simultaneous STT requests, and active agent slots; a
  single TTS WebSocket connection can multiplex "dozens" of concurrent
  generation contexts within its plan's TTS concurrency ceiling. Usage
  above the plan's monthly credit allowance is billed or blocked depending
  on account configuration.
notes: >-
  Cartesia does not publish numeric requests-per-minute or requests-per-day
  ceilings; the governing dimension is concurrent open connections/requests
  per plan. Verify current per-plan concurrency numbers on the pricing page
  during reconciliation, since they are tied to commercial tiers that can
  change.
sources:
- https://cartesia.ai/pricing
- https://docs.cartesia.ai/api-reference/tts/websocket
- https://docs.cartesia.ai/api-reference/stt/websocket
responseCodes:
  throttled: 429
limits:
- name: Concurrent TTS Requests (Free)
  scope: account
  metric: connections
  limit: 2
  notes: Applies to /tts/bytes, /tts/sse, and /tts/websocket combined.
- name: Concurrent TTS Requests (Pro)
  scope: account
  metric: connections
  limit: 3
  notes: Applies to /tts/bytes, /tts/sse, and /tts/websocket combined.
- name: Concurrent TTS Requests (Startup)
  scope: account
  metric: connections
  limit: 5
  notes: Applies to /tts/bytes, /tts/sse, and /tts/websocket combined.
- name: Concurrent TTS Requests (Scale)
  scope: account
  metric: connections
  limit: 15
  notes: Applies to /tts/bytes, /tts/sse, and /tts/websocket combined.
- name: Concurrent STT Requests (Free)
  scope: account
  metric: connections
  limit: 8
  notes: Applies to /stt, /stt/websocket, and /stt/turns/websocket combined.
- name: Concurrent STT Requests (Pro)
  scope: account
  metric: connections
  limit: 12
  notes: Applies to /stt, /stt/websocket, and /stt/turns/websocket combined.
- name: Concurrent STT Requests (Startup)
  scope: account
  metric: connections
  limit: 20
  notes: Applies to /stt, /stt/websocket, and /stt/turns/websocket combined.
- name: Concurrent STT Requests (Scale)
  scope: account
  metric: connections
  limit: 60
  notes: Applies to /stt, /stt/websocket, and /stt/turns/websocket combined.
- name: Agent Slots (Free)
  scope: account
  metric: agents
  limit: 1
  notes: Number of configured voice agents allowed simultaneously.
- name: Agent Slots (Pro)
  scope: account
  metric: agents
  limit: 3
  notes: Number of configured voice agents allowed simultaneously.
- name: Agent Slots (Startup)
  scope: account
  metric: agents
  limit: 5
  notes: Number of configured voice agents allowed simultaneously.
- name: Agent Slots (Scale)
  scope: account
  metric: agents
  limit: 10
  notes: Number of configured voice agents allowed simultaneously.
- name: TTS WebSocket Context Multiplexing
  scope: connection
  metric: contexts
  limit: dozens per connection (published guidance, not a hard number)
  notes: A single wss://api.cartesia.ai/tts/websocket connection can run many concurrent context_id generations; still bounded by the account's overall concurrent TTS request limit.
- name: STT WebSocket Audio Cadence
  scope: connection
  metric: frames
  limit: 100ms recommended chunk size
  notes: Binary audio frames should be sent in roughly 100ms chunks matching the negotiated encoding and sample_rate.
- name: Access Token Lifetime
  scope: token
  metric: seconds
  limit: 3600
  notes: Short-lived scoped tokens minted via POST /access-token expire after at most 3600 seconds.
policies:
- name: Concurrency Over Rate
  description: Cartesia governs load primarily through simultaneous connection/request ceilings per plan rather than a rolling per-minute request counter.
- name: Enterprise Custom Concurrency
  description: Enterprise agreements can raise TTS/STT concurrency and agent slot limits beyond the Scale tier.
- name: Backoff Strategy
  description: Clients should implement exponential backoff with jitter and honor Retry-After on 429 responses, and should reuse a single WebSocket connection across many contexts rather than opening one per generation.
maintainers:
- FN: Kin Lane
  email: kin@apievangelist.com