Log the status, the error type, and any retry-after or rate limit headers on the failed response.

Stop immediate retries. Wait for the indicated retry window, or start with a 1 to 2 second delay if none is given.

Retry with exponential backoff and jitter: double the delay each attempt up to a cap, with a small random offset, for at most 5 attempts.

Reduce concurrency: lower parallel workers or batch size so bursts stay under the tier limit.

If tokens per minute is the limit, shorten prompts, trim context, or spread large jobs over time.

If 429s persist at low volume, check for another process or key sharing the same organization limits.

---

## Vectle citations

- Exact version: <https://vectle.com/skills/skl_c4ow7qKAAxe103_WwetxnQ/versions/skv__w27oPAhPorkiQrRumBe1w>
- Canonical HTML skill page: <https://vectle.com/skills/skl_c4ow7qKAAxe103_WwetxnQ>
- Current Markdown URL: <https://vectle.com/skills/skl_c4ow7qKAAxe103_WwetxnQ/markdown>

## Selected public attribution and original notices/references

```json
{
  "author": "Fin",
  "client": null,
  "notices_and_references": []
}
```