Limit changes

Updated Claude 3.5 Haiku Google Cloud availability note

Detected October 23, 2026, 05:15 UTC · Claude API rate limits

The rate limits documentation previously stated that Claude 3.5 Haiku was retired except across Google Cloud generally. It now clarifies that the retirement exemption applies specifically to Google-managed Claude model deployments on that platform.

Anthropic’s rate limits reference outlines capacity thresholds, service tiers, and supported cloud integrations for developers using the API. In the updated model table, the status notation for Claude 3.5 Haiku was revised. Where the page previously noted that the model was retired everywhere except generally across Google Cloud, it now specifies that the retirement exception is restricted to Google-managed Claude model services on that platform. Additionally, the page navigation was updated to incorporate a link for Google Cloud identity and access management permissions alongside existing AWS settings. For engineering teams, this edit narrows the official scope of where Claude 3.5 Haiku remains operational. Developers provisioning workloads directly through standard API keys continue to find the model retired, whereas organizations accessing Claude through Google's first-party managed infrastructure retain access under their cloud platform allocations. The documentation does not clarify whether customer-managed installations on Google Cloud previously had access that is now revoked, or if this revision simply represents a wording correction. Furthermore, the page does not provide numerical throughput quotas or an end-of-life schedule specifically for the Google-managed deployment of this model.

We compare this page once a day and record every difference. The text above is written by Claude Network from that comparison; none of it is copied from the source.

Source: Claude API rate limits