Claude Pro usage limits (2026): what you get and when it runs out

Short answer: Claude Pro is metered by two independent clocks — a rolling 5-hour window and a weekly cap. That is the same mechanism used across paid tiers; what Pro buys you is a particular bucket size, not a different system. Most people who feel Pro is too small are not doing more work than the plan supports. They are sending more than they realise.

The two clocks

1. A rolling 5-hour window. Your first message starts a session, and everything in the next five hours draws from the same short-term allowance. It refills continuously rather than resetting at a fixed clock time, which is why the moment you regain room moves depending on when you started.
2. A weekly cap on a fixed 7-day cycle, governing total volume regardless of how you spread it.

They are independent meters. Waiting out the window does nothing for a spent weekly cap — the single most common misunderstanding on this plan. The window mechanics are covered in depth on our 5-hour limit page, and the weekly cycle on Claude’s weekly limit.

We are not going to print exact allowance numbers. Anthropic has revised them more than once, and a stale figure on a limits page gets quoted back as fact months later. Check your current plan details.

Why agent work drains Pro faster than chat

This is the part that surprises people upgrading from chat use to Claude Code.

In a chat, roughly what you type is what you send. In an agent session, what you type is a small fraction of it. The agent reads files to orient itself, and re-reads many of them on later turns. It runs commands and pulls their output — including long build logs and test runs — into context. It carries conversation history forward even when only the last few exchanges are relevant.

None of that is visible as effort. It is the reason a one-line instruction in a large repository can cost more than an afternoon of chat, and why Pro can feel generous one week and impossibly tight the next depending on what you were working on. Where that burn concentrates is covered in what Claude Code costs you.

Pro versus Max

Max raises the ceiling on both clocks. It does not remove them, and it does not change the mechanism — heavy agent sessions can still exhaust a Max allowance, just later. Our Claude Max limits page covers what the higher tiers actually change.

The decision is worth making on evidence rather than frustration:

• If you exhaust Pro mid-week doing work that genuinely needed doing, Max is the direct fix and you should buy it.
• If a large share of your burn turns out to be repeated reads and noisy command output, you would be buying headroom to keep wasting at a larger scale.

Both outcomes are common. The difference between them is measurable, and it takes about a minute to check.

What to do when you hit the limit

In the short term: wait out the window, or shift to lighter work that does not need the agent. Our usage limit reached page covers the immediate options. For the wider picture of how limits work across Claude Code, start at Claude Code usage limits.

In the medium term, measure. The free hypermile-audit CLI reads your local Claude Code transcripts and reports where the tokens actually went, per session and per project. It runs on your machine and uploads nothing.

Then decide with numbers. If the waste is real, Hypermile forecasts your cap before you hit it and trims the noisy parts of each prompt so Pro lasts longer — download it here.