Amazon Bedrock adds explicit prompt caching for GPT-5.6 models

AWS has made OpenAI’s GPT-5.6 Sol, Terra, and Luna generally available on Amazon Bedrock and added explicit prompt caching for those models. The feature lets developers choose which parts of a prompt should be cached and reused across requests.

Prompt caching matters most in agentic workflows, where the same system instructions, tool definitions, and reference documents may be sent repeatedly. AWS says cached input is billed at a 90 percent discount and remains reusable for 30 minutes.

The GPT-5.6 family is served through an OpenAI-compatible Responses API on Bedrock’s bedrock-mantle endpoint. Sol is positioned for complex reasoning and agentic coding, Terra for balanced production use, and Luna for high-volume work such as classification and summarization.

The release gives AWS customers access to newer OpenAI models while keeping Bedrock governance and security controls. The practical benefit depends on workload shape: applications with large repeated context should gain more from explicit caching than one-off prompts.