Claude API Rate Limit Exceeded: Fixes

A 429 rate_limit_error means your organisation exceeded one of several limits: requests per minute, input tokens per minute, or output tokens per minute.

Read the response headers

Anthropic returns anthropic-ratelimit-requests-remaining, the token equivalents, and a retry-after value. Respect retry-after literally instead of guessing.

Identify which limit you hit

The error message names the limit. Token-per-minute limits are hit far more often than request limits when you send long documents, so measure tokens, not calls.

Add client-side throttling

A token-bucket limiter in front of your API client smooths bursts. Cap concurrency to a fixed worker pool rather than firing requests in parallel from every user session.

Use prompt caching

Caching a long, stable system prompt dramatically reduces billed input tokens per minute and keeps you inside ITPM limits.

Raise your tier

Rate limits scale with usage tier. Sustained spend automatically advances tiers; enterprise volumes can be raised by contacting Anthropic sales.

If the errors are 529 instead of 429, the problem is Anthropic-side capacity — see the 529 overloaded guide. If your key stopped working entirely, see API key authentication failing.