Guides de la plateforme
Errors and rate limits
Troubleshoot failures using status codes and request context.
Common status codes
| Status | Action |
|---|---|
| 401 | Check that the key is complete, enabled, and not expired. |
| 403 | Check model, group, or IP restrictions. |
| 404 | Check the base URL, path, and model name. |
| 429 | Reduce concurrency and retry with backoff. |
| 5xx | Record request details and retry. Submit a ticket if failures persist. |
Rate limits
Exact limits depend on the account, model, group, and upstream channel. Do not assume fixed RPM or TPM. Use the response, current policy, and Console logs.
Retry strategy
Use exponential backoff only for retryable 429 and temporary 5xx responses.
Set a retry limit and retain request correlation details.
Do not automatically retry authentication, permission, or invalid-parameter errors.