OpenAI API error
OpenAI API 429 rate_limit_error slow_down: what it means and how to fix it
429 rate_limit_error slow_downRetry? YesLast verified 3 documented fixes2 sources
What it means
Your request rate increased faster than the service can safely handle. It can occur even when traffic is within its RPM and TPM limits, because it reflects how quickly traffic increased.
Causes
- Ramping traffic too quickly; as a rule of thumb, once traffic reaches 1 million input TPM, increase it by no more than 50% every 15 minutes
- On endpoints that previously returned 503 slow_down, rapid traffic increases now return 429 slow_down
How to fix it
- Fix 1.
Follow the Retry-After header when present; if missing, increase the delay between retries and add a small random delay.
- Fix 2.
Reduce your request rate, then increase it gradually, and keep the traffic pattern steady.
- Fix 3.
Handle both 429 (RateLimitError) and 503 (InternalServerError) in SDK error handlers and inspect error.code.
More OpenAI API errors
All 23 OpenAI API errors400 invalid_request_error (Invalid service_tier argument)Retry? Not stated401 Invalid AuthenticationRetry? Not stated401 Incorrect API key providedRetry? Not stated401 You must be a member of an organization to use the APIRetry? Not stated401 IP not authorizedRetry? Not stated403 Country, region, or territory not supportedRetry? Not stated429 credit_balance_exhaustedRetry? No429 Rate limit reached for requestsRetry? Yes
HTTP 429 at other providers
Compare 429 across providers429 ThrottlingExceptionRetry? YesAmazon Bedrock429 ModelNotReadyExceptionRetry? YesAmazon Bedrock429 rate_limit_errorRetry? YesAnthropic API429 rate_limit_error (enforced_spend_limit_reached)Retry? NoAnthropic API429 Too Many Requests (Rate limit is exceeded)Retry? YesAzure OpenAI429 Too Many Requests (System is experiencing high demand)Retry? YesAzure OpenAI429 Too Many Requests (temporary rate limit adjustment)Retry? YesAzure OpenAI429 Too Many Requests (token budget exceeded by request parameters)Retry? Not statedAzure OpenAI429 Too Many RequestsRetry? YesCohere API429 - Rate Limit ReachedRetry? Not statedDeepSeek API429 rate_limit_exceededRetry? YesGemini API429 quota_exceededRetry? YesGemini API
FAQ
What does the OpenAI API error "429 rate_limit_error slow_down" mean?
Your request rate increased faster than the service can safely handle. It can occur even when traffic is within its RPM and TPM limits, because it reflects how quickly traffic increased.
How do I fix OpenAI API "429 rate_limit_error slow_down"?
There are 3 documented fixes. Start with the first: Follow the Retry-After header when present; if missing, increase the delay between retries and add a small random delay.
Should I retry OpenAI API "429 rate_limit_error slow_down"?
Yes. The official docs treat this error as retryable.
Which other AI APIs return HTTP 429?
Per their official error docs: Amazon Bedrock (429 ThrottlingException; 429 ModelNotReadyException), Anthropic API (429 rate_limit_error; 429 rate_limit_error (enforced_spend_limit_reached)), Azure OpenAI (429 Too Many Requests (Rate limit is exceeded); 429 Too Many Requests (System is experiencing high demand); 429 Too Many Requests (temporary rate limit adjustment); 429 Too Many Requests (token budget exceeded by request parameters)), Cohere API (429 Too Many Requests), DeepSeek API (429 - Rate Limit Reached), Gemini API (429 rate_limit_exceeded; 429 quota_exceeded; 429 too_many_requests; 429 RESOURCE_EXHAUSTED), Groq API (429 Too Many Requests), Mistral API (429 Too many requests), OpenRouter (429 rate_limit_exceeded), Vertex AI (429 RESOURCE_EXHAUSTED; 429 Resource exhausted, please try again later; 429 Too many requests. Exceeded the Provisioned Throughput), xAI API (429 Too Many Requests).
Sources
Last verified against 2 sources: OpenAI API's official docs and official SDK repositories.
- developers.openai.comhttps://developers.openai.com/api/docs/guides/error-codes
- developers.openai.comhttps://developers.openai.com/api/docs/guides/rate-limits
Newsletter
Know when the APIs change
One email per week. New model releases, API changes, what broke, and how developers are working around it.