How to Implement Prompt Caching on Amazon Bedrock and Cut Inference Costs in Half

Add one line of code to stop reprocessing static system prompts on every API call. Real benchmarks across 3 Nova models — 49% savings with caching alone, 97% when combined with model selection.