Errors and rate limits
What each status code means, and which ones are worth retrying.
Two limits apply per key: requests per minute (RPM) and parallel requests (concurrency). Both come from your pack settings.
| Pack category | RPM / key | Concurrent / key | Context limit |
|---|---|---|---|
| VIP | 60–140 | 5–30 | 524K |
| Unlimited (regular) | 30 | 3 | 256K |
| Token | 60 | 5 | unlimited |
The table is generated from the packs on sale right now, so it always matches what the gateway enforces. Your own limits are also shown on the pack card and on your API Key page.
The gateway also enforces a global cap on active requests. Beyond it, requests are answered with 503 gateway_busy.
Every rejection carries a machine-readable code alongside the HTTP status.
| Status | Meaning | What to do |
|---|---|---|
400 | invalid_body or invalid_model. The JSON is malformed or the model field is missing. | Check the payload and the model name. |
401 | invalid_api_key. The key is invalid, revoked, or expired. | Check the key, or reveal it again from the dashboard. |
402 | quota_exhausted. The token balance is empty and no other pack is active. | Buy a pack to continue. |
403 | model_not_allowed, or key_too_many_locations. | Check your pack's model list, or stop using the key from too many devices. |
429 | rate_limited or concurrency_limited. RPM or the parallel-request limit was hit. | Wait a moment and retry, honouring Retry-After. |
502 | upstream_unavailable or upstream_error. The provider could not be reached or returned an unexpected error. | Retry shortly. Internal details are never exposed. |
503 | gateway_busy or maintenance. The global cap is saturated, or maintenance is running. | Retry shortly. Announcements go out on Telegram. |
429 means you hit your own limit. 503 means the provider or the gateway is having trouble right now. They are not the same thing, and the fix is different.
The only message you will ever receive for a hiccup on the OAO side is a short public message plus a request_id:
{
"error": {
"code": "upstream_error",
"message": "Upstream request failed.",
"request_id": "req_20261103_ab12cd"
}
}Wait a moment and retry. If it persists, send the request_id to support on Telegram for a quick trace.
Retry 429 and 503 with exponential backoff, and always honour the Retry-After header when it is present. Do not retry 400, 401, 402 or 403, because the request will fail the same way every time.
For the questions people ask most often: FAQ

