API Rate Limits Updated for Haiku Availability and Deprecations

This documentation page specifies traffic caps and rate limits across various Claude models and endpoints. The recent revision introduces Claude Haiku 5.5 to the rate limit structure while formalizing deprecation statuses and adjusting legacy model availability across third-party platforms. Specifically, Claude Haiku 3.5 previously remained accessible on both Amazon Bedrock and Google Cloud following general retirement; the documentation now lists it as retired on Amazon Bedrock, maintaining an exception only for Google Cloud. Alongside this change, the text completion endpoint and Claude Sonnet 4.5 have both received deprecation labels, with Sonnet 4.5 still sharing an aggregate quota pool with version 4.6.

In practice, developers relying on legacy text completions or older Sonnet 4.5 pipelines must account for their deprecated standing and prepare workloads for updated alternatives. For teams consuming models via cloud hyperscalers, organizations hosted on Amazon Bedrock can no longer rely on Haiku 3.5 according to the revised status notes. However, the documentation excerpt does not disclose specific numeric request-per-minute or token-per-minute thresholds for the newly added Haiku 5.5 tiers, nor does it specify exact shutdown schedules for the endpoints and models marked as deprecated.