429 Too Many Requests — what it means & how to fix
429 Too Many Requests
HTTP 429 means the API is rejecting requests because you’ve exceeded a limit, usually requests-per-minute or tokens-per-minute, sometimes quota. The fix is to back off, retry, and reduce burst load.
What causes it
- Too many requests or tokens per minute for your tier.
- Parallelism or retry storms multiplying your call rate.
- Shared keys: multiple apps or teammates using the same key at once.
How to fix it
- Implement exponential backoff with jitter; honor Retry-After.
- Cap concurrency and queue non-urgent calls.
- Use separate keys per app so one workload can’t starve another.
- Upgrade your tier for higher limits.
Rate limits by provider
LLM API rate limits compared →
OpenAI, Anthropic and Gemini tiers side by side, and how each one handles a 429.
Keep working on a fixed monthly plan
Standard Compute gives each key a high request ceiling (300 requests a minute on Starter, 2,000 on Pro) and retries transient upstream 429s, falling back to another provider for the same model where one exists. A client that bursts past its key’s ceiling still gets a 429, so keep backoff in place.
Plans from $19 a month with $19 of compute. Eligible new accounts can try $0.25 of compute first, no card. Trial terms
FAQ
Should I just retry on 429?
Retry with exponential backoff, not immediately. Hammering the endpoint makes it worse. Respect the Retry-After header when the provider sends one.