Heads up: headless claude -p eats your 5-hour limit about 1.5x faster than using Claude Code normally
retained text
**TL;DR:** If you run Claude Code headless (`claude -p` in scripts or pipelines), it burns through your 5-hour window about 1.5 times as fast as the same amount of normal, interactive use. I measured it on four Max 20x accounts. **What I did** I'm building a usage tracker. Every couple of minutes it reads the 5-hour and weekly usage meters on four Max accounts. It matches them against how many tokens each message used (Claude Code keeps those counts in local log files). That lets me compare stretches where I was mostly running headless with stretches where I was mostly working interactively. **What I found** On the three accounts that run my headless pipelines, for every 1% of the weekly limit used: * mostly interactive work used about **4.4%** of the 5-hour window * mostly headless work used about **7%** **Image 2** shows this. Across all four accounts, once you account for which account it was and which week, headless comes out at **1.49x**. The true figure is very likely somewhere between 1.4x and 1.6x. Each account on its own indicates the same thing (**image 3**). My main account barely runs anything headless, so its estimate is very uncertain. **Could it be something else?** I tested the obvious explanations on the three accounts that do headless work, and none of them makes it go away: * **Cache settings:** headless and normal sessions cache the same way. Comparing like with like still gives about 1.46x. * **Start-up cost of each run:** counting it in still leaves about 1.45x. * **The meters updating late:** shifting the timing 10 to 30 minutes either way only lowers it a little, to 1.42x at worst. * **Running lots at once, different models, effort settings, time of day:** checked one at a time, it stays between about 1.5x and 1.6x. All of them together gives 1.52x. **What it means for you** If you keep hitting the 5-hour limit while your weekly limit still has room, and you use `claude -p` a lot, this may be part of the reason. **The fine print** * It's my own accounts and workloads. One account did most of the heavy headless work, and leaving it out still gives 1.50x. * Max 20x only. I can't say anything about Pro or Max 5x. * I can show that it happens, not why. Full write-up, every check, plus the data and scripts to re-run it yourself: [https://github.com/jonathanavis96/claude-usage-tracker/blob/main/docs/findings/2026-10-05-headless-repro/PROOF.md](https://github.com/jonathanavis96/claude-usage-tracker/blob/main/docs/findings/2026-10-05-headless-repro/PROOF.md) **The tracker** This came out of a usage tracker I'm building: [https://alldonesites.com/claude-usage-tracker/](https://alldonesites.com/claude-usage-tracker/) It's still a work in progress, so I'll do a separate post later explaining how it all works. The **first image** is a sneak preview. It’s already picking up something I haven’t seen anyone mention: since about **9 Oct, the 5-hour window looks about 15% bigger**, while the **weekly limit has barely moved**. So if you max out every window, you now hit the weekly cap after about 9% fewer of them. All four accounts show it. It’s still provisional, so treat it as a maybe until it’s confirmed.
document fetched 2026-10-24 10:41 · canonical https://reddit.com/r/ClaudeCode/comments/1xdk3cn/heads_up_headless_claude_p_eats_your_5hour_limit