Operations

Rate limits

Per-key limits by plan (60, 180 or 600 requests a minute), concurrency and word throughput, and how to back off.

Limits apply per API key, so a runaway batch job on one key can't starve your production traffic on another. They follow your plan.

Limits by plan

Plan	Requests / min	Concurrent requests	Words / min	Async jobs in flight	Voices
Basic API	60	5	150k	50	3
Pro API	180	20	600k	500	15
Ultra API	600	50	2M	5,000	Unlimited
Test keys (`sk_test_`)	60	5	50k	20	Unlimited

Deep strength and calibrated Detect need Pro API or Ultra API (403 permission_denied otherwise). Queued jobs run in plan order: Ultra API first.

Rate limit headers

Every response tells you where you stand, so you can slow down before you hit the wall:

Header	Meaning
`X-RateLimit-Limit`	Requests allowed in the current one-minute window.
`X-RateLimit-Remaining`	Requests left in the window.
`X-RateLimit-Reset`	Unix time when the window resets.
`X-WordLimit-Remaining`	Words left in the current word-throughput window.
`Retry-After`	Seconds to wait. Only on `429`.

Handling 429

Respect Retry-After when present; otherwise back off exponentially with jitter. Rejected requests are never billed.

async function withRetry(call: () => Promise<Response>, max = 5) {
  for (let attempt = 0; ; attempt++) {
    const res = await call();
    if (res.status !== 429 && res.status < 500) return res;
    if (attempt >= max) return res;
    const retryAfter = Number(res.headers.get("retry-after"));
    const wait = retryAfter > 0 ? retryAfter * 1000 : Math.min(2 ** attempt * 500, 20000) * (0.5 + Math.random());
    await new Promise((r) => setTimeout(r, wait));
  }
}

For large batches, use async jobs: they queue on our side instead of competing for your per-minute request budget.