---
title: "Prompt caching: 10x cheaper LLM tokens, but how?"
description: "Jonathon Kitchen explains the transformer state behind prompt caching. A token’s cost depends on where it appears and whether the provider has seen the prefix before."
author: "Pau Friedman"
kind: "link"
published_at: "2025-12-19"
url: "https://pau.fm/blog/2025/12/19/prompt-caching-10x-cheaper-llm-tokens-but-how"
markdown_url: "https://pau.fm/blog/2025/12/19/prompt-caching-10x-cheaper-llm-tokens-but-how.md"
canonical_url: "https://pau.fm/blog/2025/12/19/prompt-caching-10x-cheaper-llm-tokens-but-how"
external_url: "https://ngrok.com/blog/prompt-caching/"
---

# [Prompt caching\: 10x cheaper LLM tokens\, but how\?](<https://ngrok.com/blog/prompt-caching/>)

API pricing makes a token look like a standard unit, but the compute underneath it is path
dependent. A repeated prompt prefix can be ten times cheaper because the provider reuses its
attention state, while the same text in a different position has to be computed again.

“The token” is an accounting abstraction over GPU work whose actual cost depends on sequence,
position, and recurrence. Useful commodities become standardized at the interface long before their
production becomes uniform.