Rate limits

API providers limit usage to share capacity fairly. Limits are usually expressed as requests per minute and tokens per minute, sometimes split into input and output tokens, and they often rise as your account's usage tier increases. Check your provider's console for your current limits.

When you exceed a limit, the API returns an HTTP 429 error. The standard response is to retry with exponential backoff and jitter, and to honor any retry-after header the response includes. The official SDKs retry some of these errors automatically.

Agents hit rate limits more easily than chat apps because one task can make dozens of calls, and parallel agents multiply that. Prompt caching, smaller models for simple steps, batch processing for offline work, and a queue that controls concurrency all help you stay within limits.