Rate limits
Per-org per-minute request caps that scale with your plan. Plus per-key monthly spend caps as a hard rail (see Spend caps).
Per-plan RPM caps
| Plan | RPM cap | Typical headroom |
|---|---|---|
| Free | 20 req/min/org | Hobby + prototype |
| Pro | 60 req/min/org | Small production workloads |
| Enterprise | 240 req/min/org + custom | Production at scale; contact us for higher. |
The cap is org-wide, not per-key, so 5 keys on a Pro plan share the 60 RPM pool. Limits apply only to /v1/* calls — admin / dashboard endpoints aren't counted.
Check your headroom
Every successful /v1/chat/completions response includes:
X-RateLimit-Remaining-Plan: 17
X-RateLimit-Reset-Plan: 1781449320 # unix seconds when the per-minute counter resets When you exceed
HTTP 429 with:
{
"error": {
"message": "Rate limit exceeded for plan 'free'. Upgrade or slow down.",
"type": "rate_limit_exceeded"
}
} Back off until X-RateLimit-Reset-Plan. The playground surfaces a friendly banner with a 60s countdown when this happens — your own client should do similar.
Why we cap
So one customer's runaway loop can't degrade the upstream-provider experience for everyone else routed through the same shared upstream pool. Pro+ plans get a larger share; Enterprise can negotiate dedicated upstream allocation.