Skip to Content
Rate limits

Rate Limits & Quotas

Each API key is subject to per-minute request limits and concurrency caps to ensure fair use and stable service for all customers. This page documents the fixed input caps and how to handle 429 rate_limited responses.

Per-request input caps

These hard caps are enforced on every request. Exceeding the text cap returns 400 validation_error; oversized or over-length audio returns 413 validation_error (or 400 validation_error when the over-length duration is detected by probing).

EndpointMaximum input per request
POST /v1/tts500 characters of text
POST /v1/stt50 MB file / 30 minutes of audio

Request-rate & concurrency limits

Per-minute request limits and concurrency caps are enforced per API key, not per account or IP. Unlike the input caps above, these values are policy-driven — they are set server-side per key and can change without a redeploy — so no fixed public number is published here. When you exceed your request-rate limit the API returns 429 rate_limited; when too many requests are in flight at once it returns 429 concurrency_limit_exceeded. Calls to a free model are additionally capped by its 60-minute monthly allowance (429 quota_exceeded) — see Free allowances. Need higher limits for your key? Contact api@revolab.ai.

Discovery endpoints (GET /v1/voices, GET /v1/models) have their own generous, non-billable request bucket — normal SDK usage (listing once per session) never sees it; only a runaway request loop is throttled.

Note that requests rejected with validation errors (4xx) still count toward your request-rate window — the limiter runs before validation, by design, so malformed traffic cannot bypass it. Only the rate window is consumed; nothing is billed.

Rate limit response

When a limit is exceeded, the API returns HTTP 429 with the standard /v1/ error envelope:

{
  "error": {
    "code": "rate_limited",
    "message": "Rate limit exceeded.",
    "request_id": "req_01HXYZ...",
    "retry_after": 12
  }
}

The Retry-After HTTP header is also set with the same value in seconds.

To back off before hitting a 429, read the rate-limit headers returned on every billable response:

  • X-RateLimit-Limit — the maximum requests allowed in the current window.
  • X-RateLimit-Remaining — requests remaining in the current window.
  • X-RateLimit-Reset — seconds until the window resets.

Retry & exponential back-off

We recommend exponential back-off with jitter when handling 429 responses:

  1. On first 429: wait retry_after seconds from the response body.
  2. On subsequent 429s: double the wait time each retry (cap at 64 s).
  3. Jitter: add a random 0–500 ms to each wait to avoid thundering herd.
  4. After 5 retries with no success, surface the error to the user.

Higher limits

If your use case requires higher RPM, concurrent requests, or text length limits, contact api@revolab.ai.

  • Errors — full error catalog including rate_limited
  • Models — compare STT and TTS models
ErrorsOpenAI