Reusing pre-processed input across requests to cut cost and latency.
1 min readupdated 2026-07-04
/ quick answer
Anthropic and OpenAI cache expensive prefixes (long system prompts, docs) between calls. Cached tokens are ~10x cheaper and skip re-encoding — key for RAG and agent loops. Reusing pre-processed input across requests to cut cost and latency.
Reusing pre-processed input across requests to cut cost and latency. Anthropic and OpenAI cache expensive prefixes (long system prompts, docs) between calls. Cached tokens are ~10x cheaper and skip re-encoding — key for RAG and agent loops. In practice: A 50k-token system prompt costs $0.05 fresh but $0.005 cached across the next 5 minutes. This dictionary node is part of the Onexial knowledge graph and links to related concepts, workflows and tools below.
Definition
Anthropic and OpenAI cache expensive prefixes (long system prompts, docs) between calls. Cached tokens are ~10x cheaper and skip re-encoding — key for RAG and agent loops.
Example
A 50k-token system prompt costs $0.05 fresh but $0.005 cached across the next 5 minutes.
/ frequently asked
What is Context Caching?
Anthropic and OpenAI cache expensive prefixes (long system prompts, docs) between calls. Cached tokens are ~10x cheaper and skip re-encoding — key for RAG and agent loops.
What is an example of Context Caching?
A 50k-token system prompt costs $0.05 fresh but $0.005 cached across the next 5 minutes.
Why does Context Caching matter for AI and automation?
Reusing pre-processed input across requests to cut cost and latency. It connects to the workflows, prompts and tool stacks linked on this page, so you can move from definition to execution without leaving Onexial.