Home › Blog › LLM 429 Errors and Retries: How to Stop Paying Twice
ReliabilityLLM 429 Errors and Retries: How to Stop Paying Twice
Retries are necessary, but careless ones waste money: a timed-out request may still finish and bill you, and a retry storm multiplies spend.
Quick answer
Use exponential backoff, sensible timeouts, a queue for bursts, and load balancing across routes, and never retry errors that can't succeed.
What each error means
- 429: rate or usage limit. Back off and retry, or spread load.
- 5xx: provider or gateway trouble. Retry with backoff, or fail over.
- 400, 401, 403, 404: retrying won't help. Fix the request.
Where money leaks
- Client timeouts shorter than real response times, so the first request still bills and the retry bills again.
- Retry loops without a cap, multiplying traffic during an outage.
- Several layers each retrying: SDK, app and queue.
The fix
- Retry in one layer only, with exponential backoff and a cap.
- Set timeouts above your real p99 latency.
- Queue bursty background work instead of firing it all at once.
- Load balance and fail over across healthy routes so fewer requests hit limits.
See errors and retries and reliability.
Frequently asked questions
Do failed LLM requests cost money?
Requests that error before processing usually don't, but a request that times out on your side can still complete and bill.
How should I handle OpenAI 429 errors?
Back off exponentially, cap retries, queue bursts, and spread load across routes.
Want us to find your 20%?
Book a free 15-minute API cost teardown. Bring last month's provider usage and we'll show you your biggest waste and the first fix we'd ship.
Get a Free 15-Minute API Cost Teardown