How Amazon Bedrock prompt caching cuts AI costs

Prompt caching stores a snapshot of the processed input so later requests with the same prefix skip redundant work, cutting input token costs by up to 90 percent on cache hits while keeping model quality unchanged.