Claude Code 5-Hour vs Weekly Limit: Why Usage Burns Faster

Two people can both use a full 5-hour window and still end the week at very different percentages. The session limit measures time; the weekly limit is filled by what happens inside that time. Here is how the two limits relate and where the extra usage usually comes from.

The two limits, in plain terms

On a Claude subscription, usage is counted in a rolling session window of about five hours that starts with your first message. When that window is used up, Claude Code tells you when it resets.

Paid plans also have a weekly limit on top of the session window. The weekly limit does not reset every five hours; it covers a longer period. That is why a good day can still leave you with a small weekly allowance, and why the same five hours can feel cheap on Monday and expensive on Friday.

Check where you stand with /usage

Run /usage to see your current usage. Claude Code also shows the reset time when you hit a limit, so you can decide whether to wait, switch to another agent, or stop for the day.

Checking before a long session is more useful than checking after it. If the weekly number is already high, a big refactor with a large context is a bad place to spend the rest.

Why the same 5 hours can cost different weekly percentages

There is no exact formula that turns one session window into a weekly percentage. The two limits are related, but they are not a simple count of messages or minutes. These are the likely factors:

• Model. Opus uses more of your limit than Sonnet for the same work. A heavy model on routine edits is an expensive habit.
• Context size. A long conversation re-sends its history on every message. Much of that is read from cache, but cache reads still count, so a big context makes every turn more expensive.
• Effort level. More thinking means more tokens. High effort on a task that only needs a rename often spends usage without improving the result.
• Tool output size. Long test logs, large file reads and big diffs enter the context and stay there for later turns.
• Plan type. Different plans have different allowances, so the same work can map to a different percentage from one account to another.

This is why comparing raw hours is misleading. Two people can both sit in Claude Code for five hours; one is editing a small file with a light model, the other is reading logs into a very large conversation with the top model. The second person fills the weekly limit much faster.

How to protect the weekly limit

• Compact earlier. Set auto-compact to a 400K to 500K window so the conversation does not grow to the default threshold before anything is summarized. Our auto-compact guide shows the command, the flag and the settings key.
• Match effort to the task. Use --effort <level> with low or medium for routine work such as moving files, fixing text or a simple scan. Save high, xhigh or max for multi-file architecture, hard bugs and concurrency work. If you are unsure, start low and raise it only if the output is weak.
• Keep sessions short. Use /clear between unrelated tasks and /compact when a long task has a lot of stale history. A fresh context is cheaper than carrying yesterday's discussion into today's edit.
• Keep long logs out. Run noisy commands yourself, or ask for the failing lines instead of the full output. Large file reads and whole diffs add context that you pay for on every later turn.
• Give routine work to a cheaper model. Let Claude plan and review; give mechanical edits, test runs and documentation to a cheaper model or another agent. The orchestration guide covers the handoff pattern.

How Foxpit handles both limits

Foxpit is a desktop app that runs Claude Code, Codex and OpenCode as one team. It cannot add quota, and it uses your own subscriptions without reselling or working around limits. It does remove the manual work around a limit:

• Usage panel. One panel shows remaining quota and reset time for every agent that reports it, so you can see where each agent stands before you start a large task.
• Waits at the 5-hour limit. When Claude Code hits the limit, Fox parks the turn, shows when it resets and sends "continue" at the reset.
• The rest of the team keeps working. Each seat can run a different agent and model, so teammates on OpenCode or Codex can continue while Claude waits.

Related: see what to do when the Claude Code limit hits and how to set auto-compact to 400K.

Frequently asked questions