Claude/Code/Bedrock/How To/Enable Prompt Caching

• How To Enable Prompt Caching for Anthropic Models in AWS Bedrock

  • Goal

    • Ensure prompt caching is active for Anthropic Claude models on AWS Bedrock to reduce latency and token costs.

  • Preconditions

    • AWS account with Bedrock access and Anthropic Claude models enabled.
    • AWS CLI v2 configured (aws configure).
    • (Optional) Claude Code CLI installed and configured.

  • Procedure

    • 1. Configure Claude Code for Bedrock with caching enabled

      • Set the Bedrock backend:
        bash

        export CLAUDE_CODE_USE_BEDROCK=1

      • Ensure prompt caching isn't disabled:
        bash

        unset DISABLE_PROMPT_CACHING

      • (Alternative) Persist it for every session under the env key in ~/.claude/settings.json — Claude/Code/Settings:
        json

        {
          "env": {
            "CLAUDE_CODE_USE_BEDROCK": "true"
          }
        }

    • 2. Verify prompt caching via AWS CLI

      • Create a payload.json:
        json

        {
         "prompt": "Human: Test for caching.\n\nAssistant:",
         "cache_control": {"type":"default"},
         "max_tokens": 10
        }

      • Invoke the model and specify an output file (required positional outfile):
        • bash

          aws bedrock-runtime invoke-model \
           --model-id us.anthropic.claude-3-7-sonnet-20250219-v1:0 \
           --body fileb://payload.json \
           --content-type application/json \
           --region us-east-1 \
           --profile bedrock \
           response.json

        • Note: Use fileb:// (not file://) for the --body parameter to avoid base64 errors.
      • Inspect the response for caching metrics:
        bash

        jq '.CacheReadInputTokens, .CacheWriteInputTokens' response.json

        Presence of numeric values confirms prompt caching is operational.
        

    • 3. Enable caching in the Bedrock console

      • Open the Amazon Bedrock console.
      • In Playground, select an Anthropic Claude model.
      • Toggle Prompt caching to On (default for supported models).

  • Troubleshooting

    • Invalid base64 error → Use fileb:// instead of file:// for the --body parameter in your AWS CLI command.
    • Missing cache fields → Verify your account's prompt caching entitlement; contact AWS Support if unavailable.
    • cache_control not recognized → Upgrade AWS CLI/SDK to the latest version that supports cache control.
    • Console toggle absent → Confirm your region and console version support prompt caching.

  • References

    • Bedrock & Vertex · Anthropic Documentation
    • Prompt caching for faster model inference · Amazon Bedrock User Guide
    • AWS Prompt Engineering: A Developer's Guide to CLI Usage