provider_rate_limited means the model service was too busy to accept the call, on every route OpenType tried. Read this page when runs fail with this code under load.
What happened
Route:POST /v1/runs.
The model service answered that it was rate-limiting OpenType. OpenType already tried its fallback route before answering, so this code means every route it tried was limited. It is not a limit on your organization: OpenType has no request-rate limit, and your own limits are the period quotas, which answer 429 before the run starts.
Every provider_* code has the same message, “the provider call failed”. Branch on code, never on the message. The message never contains model output.
If this was the first model call, the run was refused before the model served it. You are not charged, and the hold is released, but the run stays pending: a replay with the same Idempotency-Key returns 202 with "state": "pending". If an earlier call in the same run was served (a verdict repair), the run is settled failed and a replay returns 200 with "state": "failed". In both cases, retry with a new key.
How to fix
- Retry with backoff (for example 2, 4, 8 and 16 seconds) and a new
Idempotency-Keyon each attempt. Log therequest_idof every failed attempt. - Spread bursts out. If you send many runs at once, add jitter to the backoff so retries do not arrive together.
- Do not treat this as a quota. Spend and token quotas answer
429, not503.
Example
Related
- Spend limits and quotas - the limits that do apply to your organization.
- Error handling - a status-to-action table and a retry helper for every error.
- Idempotency - when to reuse an
Idempotency-Keyand when to send a new one. - Request ids - send your own
x-request-idand quote it when you report a problem. - Problem codes - every code, its status, and whether a retry can help.