# Prompt caching doesn’t make a big CLAUDE.md free — retained record

> Retained reddit record from reddit:ClaudeCode.

Canonical: https://highsignal.app/data/records/da2b002594a48180

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeCode)  published 2026-11-07 06:55 ingested 2026-11-08 02:39

[original source ↗](https://www.reddit.com/r/ClaudeCode/comments/1xpzzde/prompt_caching_doesnt_make_a_big_claudemd_free/)

## retained text



```
Got this reply on my last post and it’s a fair point, so here’s the longer answer. Yes, Claude Code caches your session, and a cache read costs about a tenth of normal input. But three things still add up.

* The first write isn’t cheap. Your memory files get written to the cache at the start of every session, and a cache write costs more than normal input.
* Reads are cheap, not free. That same memory gets read back on every call, and a session makes a lot of calls. Every tool call counts as one.
* The cache expires. Step away long enough and the next message writes the whole session again.

So a big [CLAUDE.md](http://CLAUDE.md) gets paid for once at the higher rate, then a little on every call after that, for the whole session.

We measured this (disclosure, [minniai.com/evidence](http://minniai.com/evidence) is our site). Same 30 questions in Claude Code, two sessions a side, on a test memory before and after we split it with our hook on. Tokens went from 2,840,176 to 1,623,741, but cost only went from $0.83 to $0.57, because most of what we saved were cache reads, the cheapest kind. Other runs came in higher and lower.

So caching softens it a lot. It just doesn’t make it go away.

Do you clear between tasks or keep one long session going?
```



document fetched 2026-11-08 00:33 · canonical https://reddit.com/r/ClaudeCode/comments/1xpzzde/prompt_caching_doesnt_make_a_big_claudemd_free
