Amazon Bedrock adds prompt caching to reduce AI agent token costs
Amazon Bedrock now supports prompt caching, letting developers store a reusable prefix of an agent’s prompt—tools, instructions, and static content—and reuse it on subsequent calls. By saving the model’s key‑value state after the first prefill, later requests only process new tokens, cutting input‑token charges.
Key points
- Prompt caching stores the model’s KV state after the first request, so later calls only process new tokens
- Implicit caching cut GPT‑6 Astra input cost by about 62 percent; explicit breakpoints saved about 61 percent
- Cached prefixes require an exact text match and have a minimum 30‑minute TTL before expiration
In internal tests, a roughly 4,000‑token tools‑and‑instructions prefix wrote 4,120 tokens on the first request and read the same 4,120 tokens on later requests. Using GPT‑6 Astra, implicit caching achieved about 62 percent lower input cost, while explicit rolling breakpoints saved about 61 percent. Cached prefixes have a minimum 30‑minute TTL and are refreshed by reuse.
The blog explains both implicit (Bedrock‑chosen) and explicit (promptcachebreakpoint) modes, how to mark static sections, handle new tools, and share cached prefixes across sub‑agents. It also warns that cached prompts can leak sensitive data, so responsible AI governance is recommended.
Model pages: GPT-6 Astra → · GPT-6.1 Sol →
The story so far
2 episodes →- Amazon Bedrock adds prompt caching to reduce AI agent token coststhis story
Groundhog Day for AI agents: Prompt caching patterns for using OpenAI models on Amazon Bedrock
AWS Machine Learning Blog · 26 October 2026
Loading the full article…
This text was published by AWS Machine Learning Blog and written by Claudio Mazzoni. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- Nace AI releases Drex for Search with 92.05% on BrowseComp-Plus · 1 src
- vLLM 0.31 adds constrained decoding for 100% tool-call accuracy · 1 src
- OpenAI Agents SDK lets Python developers build web-searching, tool-using agents · 1 src
- AI agents navigate software via clicks and typing without APIs · 1 src
- Claude Code AI detects cryptominer on Mac during temperature monitor build · 1 src
Comments
via GitHub Discussions