Bedrock adds explicit caching for GPT-5.6
Amazon Bedrock now supports explicit prompt caching for OpenAI's GPT-5.6 models, with cache reads billed at a 90% discount versus uncached input tokens.
Amazon Web Services added explicit prompt caching for OpenAI’s GPT-5.6 models on Amazon Bedrock’s Responses API, alongside the implicit caching Bedrock already applies automatically. On GPT-5.6, cached-token reads are billed at a 90% discount versus uncached input tokens, while cache writes cost 1.25x the uncached rate, so AWS says a workload’s net input cost starts falling once cache reads make up roughly 20% of tokens flowing through the cache. Implicit mode requires no code changes and works once a request has a stable prefix of at least 1,024 tokens; explicit mode lets a developer mark the cache boundary directly, which AWS says typically pushes the hit rate higher for workloads like chat assistants and agentic tool loops where content after the stable prefix changes on every request.
What it means for operators
This is a straightforward cost lever for anyone running GPT-5.6 through Bedrock rather than OpenAI’s API directly: implicit caching already saves money with zero code changes, but explicit caching is worth the marginal engineering effort specifically for agentic loops and chat systems where the same system prompt or tool schema repeats across many calls. Check your cached_tokens and cache_write_tokens usage fields before deciding which mode fits your traffic pattern.