Skip to main content
Each key allows 600 requests per minute, enforced at the gateway on a fixed one-minute window. The limit applies per key on every metered route — LLM, TTS, STT, voices, pronunciation dictionaries, jobs — and at STT WebSocket upgrade. Past the limit, requests return 429 with a Retry-After: <seconds> header; retry after that many seconds.
The limit is per key, not per connection — one warm HTTP/2 connection multiplexes concurrent streams and is the recommended client pattern. Opening a connection per request adds a TCP + TLS handshake (~250 ms EU to us-west) to every call and gains you nothing.