Claude API Spend Controls Checklist

Look forA limit under the cap on the Billing page. At a limit you set, requests return HTTP 400 until you raise it or the month resets. At the tier cap they return HTTP 429 with the error code enforced_spend_limit_reached and no retry-after header, so retries fail until access resumes.

Look forA monthly spend limit with alert thresholds per workspace, set under Settings > Workspaces, and per-model request and token limits per workspace on the Rate limits page.

Look formax_turns and max_budget_usd set on Claude Agent SDK agents, and an equivalent counter in your own loop. One builder left an Agent SDK agent unattended without these settings and came back to a bill too large to say out loud. The builder now sets both, because they catch different failures.

Look forCache hit rate watched on the Usage page or pulled by workspace and API key from the Usage and Cost API, with an alert when it drops. Cache reads cost 5% of the input price on Opus 5.5 and Sonnet 5.5, 2.5% on Fable 5.1, and 10% on most other models, and on most models they do not count toward input rate limits.

Look forA budget that counts tokenizer and prompt-length effects. Claude 4.7 and later models produce about 30% more tokens for the same text, and Haiku 5.5 costs 5x more for prompts over 100,000 tokens.

Look forOn Amazon Bedrock and Google Cloud (Google managed) the provider's budgets and quotas apply. A builder whose agent ran through LiteLLM to Bedrock found budget alerts did not stop a US$37,901.73 bill. Another team found Bedrock double counting cache writes with thinking enabled.