AWS documents the real economics of Bedrock prompt caching: cache writes cost 25% more, reads 90% less, and 1-hour TTL doubles the write price

AWS Machine Learning Blog·medium signalXBlueskyLinkedInCopy link

The post gives the numbers builders actually need to decide whether caching pays. Cache write tokens are priced 25% above standard input (100% above for the 1-hour TTL option), cache reads are 90% below, and a 10,000-token document reused across 10 questions nets roughly 75% off total input cost. Minimum checkpoint size is 1,024 tokens for Claude Sonnet 4.5 and 4.6 and 4,096 for Opus; default TTL is 5 minutes, simplified cache management covers about 20 preceding content blocks per checkpoint, and multiple checkpoints must be ordered longest TTL to shortest. Time-to-first-token gains grow with prefix size and become pronounced past 10,000 tokens.