Claude Code 5-hour limit: how the rolling window works

Short answer: The Claude Code 5-hour limit is a rolling usage window on paid Claude plans. Your first message starts a session; everything in the next five hours draws from the same short-term allowance. It refills continuously — roughly five hours after that session begins — and it is not the same clock as the weekly cap. Exact room in the window varies by plan (Pro vs Max) and changes over time; verify against Anthropic’s docs. As of July 2026, the mechanics below are the stable part.

How the rolling window works

Think of the 5-hour limit as a short bucket that slides with your activity, not a calendar appointment.

1. You send a message → a session window starts.
2. Tokens you spend in that window draw from the same short-term allowance.
3. When the allowance is empty, new work pauses or throttles until the window rolls over.
4. Because it is rolling, refill is tied to when the session started, not to “3:00 pm every day.”

Heavy back-to-back agent work empties the bucket quickly. Light, spaced-out use may never notice it. That is why one developer on your team hits the 5-hour wall before lunch while another never sees it.

This page is the deep-dive on that clock. The hub for both clocks — 5-hour and weekly — is Claude Code usage limits, and the weekly clock has its own deep-dive in Claude’s weekly limit. Cross-read those two pages together if you are deciding which limit actually stopped you.

5-hour window vs weekly cap

	5-hour window	Weekly cap
Role	Short-term burst control	Multi-day ceiling
Reset	Rolling (~5 hours after session start)	Fixed 7-day account cycle
Waiting it out	Often frees you the same day	Waiting on the 5-hour clock does not help
Typical surprise	”I was fine an hour ago"	"It’s Wednesday and I’m done for the week”

You can stay under the 5-hour limit every day and still exhaust the weekly ceiling. You can also burn the 5-hour window hard on Monday morning and still have weekly room left. Knowing which message you got — and which clock the UI points at — decides whether you wait a few hours, wait for the weekly reset, upgrade, or cut burn.

Max plans use the same two clocks with larger buckets; details are in Claude Max limits.

What fills the 5-hour window fastest

The window is not a stopwatch on “hours of coding.” It is a token budget over a rolling five-hour span. The fastest drains:

• Long autonomous agent loops with many tool calls
• Build/test output pasted or captured in full
• Wide directory reads that pull unused files into context
• Re-reading the same large files across turns

A focused edit session can sit comfortably inside the window. A morning of “explore the whole repo and fix everything” often cannot.

What to do when you hit it

1. Confirm it is the 5-hour clock, not the weekly cap. If the weekly ceiling is spent, waiting ~5 hours will not restore you — see the usage limits hub.
2. Wait for the roll-over if you only need the short window. It is designed to clear within hours.
3. Upgrade only if you repeatedly exhaust even after waiting — Claude Max limits describe the larger consumer ceilings.
4. Cut waste if the same patterns keep emptying the bucket: noisy logs and repeated reads are the usual culprits.

If you are seeing HTTP 429s from the API rather than a Claude Code subscription message, that is Anthropic rate limits — a different system.

Measure before you guess

Claude Code stores local session transcripts. The free hypermile-audit CLI reads them on your machine and shows where tokens went — tool output vs file reads vs conversation — so you can see whether the 5-hour burns are real work or compressible noise. Nothing is uploaded; it is open source.

Hypermile then helps you forecast cap timing and trim noisy prompt content before it reaches the model, so the window you already have lasts longer.

Next steps

Run the audit below, then download Hypermile. For the full two-clock picture and per-plan table, go back to Claude Code usage limits. Windows-specific setup lives in the Claude Code on Windows hub.