Simplified Cache Management for Anthropic’s Claude models in Amazon Bedrock

Amazon Bedrock has updated prompt caching for Anthropic’s Claude models to improve ease of use for Claude 3.5 Haiku, Claude 3.7, and Claude 4 models.

Previously, developers needed to manage cache points manually and keep track of which cached segments should be reused. With simplified cache management, you only need to set a cache breakpoint at the end of your request. The system automatically reads from the longest previously cached prefix, which eliminates the need to manually specify which segments to reuse and reduces the effort required to manage cache logic.

By automatically identifying and applying the right cached content, simplified cache management not only helps reduce manual effort, but also helps free up more tokens since cache read tokens are not counted toward your token per minute (TPM) quotas. This can make it easier to build multi-turn workflows and research assistants, while improving both performance and cost efficiency.

Simplified cache management is available today in all regions where Anthropic Claude 3.5 Haiku, Claude 3.7, and Claude 4 models are offered on Amazon Bedrock. To get started, review the Amazon Bedrock Developer Guide and enable caching in your model invocations.