What happened
AWS has published guidance on using prompt caching in Amazon Bedrock, a feature that can reduce input token costs by as much as 90% when the same context is repeatedly sent to foundation models.
The post covers six practical prompt caching scenarios built around the Converse API: message content, system prompt, tool definition, mixed TTL, tenant isolation, and LangChain integration.
Why it matters
Repeatedly sending identical context to a foundation model is a common pattern, so caching that context could meaningfully change the cost profile of applications that rely on it.
The scenarios listed suggest the feature is meant to be applied at several layers of an application, from individual messages to system prompts and tool definitions, rather than in a single narrow way.
Because the guidance is framed around the Converse API and LangChain, it points to caching being relevant to both direct API users and teams working through a framework.
Key facts
Prompt caching in Amazon Bedrock can cut input token costs by up to 90% when the same context is repeatedly sent to foundation models.
The post walks through six prompt caching scenarios using the Converse API.
The scenarios are message content, system prompt, tool definition, mixed TTL, tenant isolation, and LangChain integration.
What to watch next
How teams apply the six scenarios in production, particularly mixed TTL and tenant isolation, which touch on cache lifetime and separation between customers.
Whether the cost savings described for repeated context hold across the different caching scenarios outlined in the post.