# I grouped 18,629 Claude Code requests by session age. After the first hour, each one carried about twice the context. — retained record

> Retained reddit record from reddit:ClaudeCode.

Canonical: https://highsignal.app/data/records/ed960171734e3a51

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


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeCode)  published 2026-11-14 02:39 ingested 2026-11-15 02:36

[original source ↗](https://www.reddit.com/r/ClaudeCode/comments/1xvr4ob/i_grouped_18629_claude_code_requests_by_session/)

## retained text



```
Last 14 days of my own logs: 139 sessions, every API response deduped by message id, and each request bucketed by how many hours had passed since its session started.

The first hour held half the requests (9,339) but only 35% of the tokens, at about 183k tokens per request. Hours 1 to 5 held 36% of the requests and 45% of the tokens, at about 326k each. After that it barely moves: about 366k per request from hour 5 to 24, and 324k past a full day.

So the growth happens early and then flattens. Only 5 of the 139 sessions stayed open past 24 hours, and everything sent after that mark came to 3% of the tokens. Most of the tokens sit in ordinary sessions that run into their second hour.

Two caveats. Hours are wall clock, so idle time counts. And nearly all of those tokens are cache reads, which bill cheaper than fresh input.

Is the flattening after a few hours compaction doing its job, or does something else cap it?
```



document fetched 2026-11-15 00:35 · canonical https://reddit.com/r/ClaudeCode/comments/1xvr4ob/i_grouped_18629_claude_code_requests_by_session
