A 429 from Claude's API can mean two very different things — a plain throughput limit, or a hard monthly spend cap — and only one of them clears itself if you just wait and retry.
Rate limit vs. spend limit — check the error body
Anthropic returns an HTTP 429 with type rate_limit_error for both a plain throughput limit and a spend cap, but the message and error_code tell them apart — a spend-cap block carries a distinct code such as enforced_spend_limit_reached. A plain rate limit response includes a retry-after header telling you how long to wait; a spend-cap block typically does not, because access pauses until the next billing cycle rather than the next minute.
If it's a plain rate limit: slow down and retry smart
Limits are measured in requests per minute, input tokens per minute, and output tokens per minute — whichever you hit first is your bottleneck. Only uncached input tokens count toward the input-token limit for most models, so prompt caching doubles as a throughput lever, not just a cost one. Back off exponentially and honor the retry-after header rather than hammering the endpoint again immediately.
If you've hit your monthly spend cap: nothing but time or a plan change fixes it
Anthropic organizations progress through usage tiers, each with its own rising monthly spend ceiling. Hitting the ceiling on your current tier blocks API access for the whole organization until the next billing cycle resets it — retrying will not help. You can request a rate limit increase from the Rate limits page in the Claude Console, or contact Anthropic support for an urgent need.