How Claude Code's 5-hour and weekly limits work

Anthropic's help centre puts it in one sentence: "The period starts when you send your first message, and your limit resets in full five hours later." So the window is not a rolling average that drains slowly. It is a block of five hours that opens on your first message, and when the block ends you get the whole allowance back. If you send one message at 9:00 and nothing else until 13:30, the window still ends at 14:00.

Claude Code shows warning messages as you approach any limit. If you reach the 5-hour limit while Claude Code is partway through a response, Anthropic says it may keep working briefly to finish the step in progress rather than stop mid-step. After that it waits. The message it shows includes when the window resets.

Which plans have one: Pro and Max do, and so do Team and Enterprise seats, where Claude Code's docs say each member's allowance "resets on a rolling five-hour window and a weekly window". Ultra is the exception: "Verified Ultra accounts don't have a five-hour limit." Anthropic announced that on 30 October 2026, starting "with most existing subscribers"; newer Ultra accounts keep a temporary limit until they are verified. The guide is the Claude 5-hour limit gone? has the detail.

The weekly limit is the one that catches people out, because it does not line up with the calendar week. Anthropic's wording: "Weekly limits apply across all models, and some plans also have a separate weekly limit for Fable. Weekly limits reset at a fixed time each week that is assigned to your account." Your reset time is yours, and you can see it under Settings > Usage on claude.ai.

Two different messages can appear when a weekly limit is hit, and they mean different things. Claude Code's docs explain that "You've hit your weekly limit" is "a seat-based usage window on a subscription plan, shared across all models, so the developer can't restore access by switching models with `/model`". A model-specific message, such as "You've hit your Opus limit", is the separate weekly limit for one model family, and switching to a model outside that family does keep you working.

The 5-hour and weekly limits run independently. You can be well inside the 5-hour window and still be stopped by the weekly one, which is the usual way a heavy week ends.

Everything you do on the account counts, on every device and in every Claude product. Anthropic's general rule is that usage depends on "the length and complexity of your conversations, the features you use, which Claude model you're chatting with, and the effort level you've selected". Its best-practices page lists the factors as message length, file attachment size, the current conversation length, tool usage, model choice, effort level, artifacts, and multi-step tasks, and adds that cached portions count less against your limits than new content.

Claude Code adds its own reasons, because it sends the whole conversation with every request. The Claude Code docs on costs list why usage climbs in a long session:

• Long context. Claude Code sends your full conversation with every request, so "a one-line question in a session that has been open all day still draws usage for the whole conversation". Cached history is read at a cheaper rate, but it is still read.
• Cache misses. Your first message after a break longer than the cache lifetime reprocesses your full context at the full rate.
• Subagents and workflows. Every subagent sends its own requests on top of the main conversation's, and "the subagent's own requests still draw on your usage". Agent teams use roughly seven times the tokens of a normal session when teammates run in plan mode.
• Scheduled tasks, cross-session messages and goal check-ins. Each one fires a new turn that sends your full context, even while the session sits idle.
• Compaction. /compact reads the conversation it summarises, so compacting a large context is itself a large request. /clear costs nothing.
• Background requests. Session recap, idle compaction and prompt suggestions can use tokens while you are idle; the docs say where to turn each one off.

The practical reading of that list is that the 5-hour window is spent mostly by context, not by the number of things you type. Five short questions in a fresh session cost far less than one short question in a session that has been open since breakfast. Running /usage in Claude Code shows a breakdown of what is behind your recent usage, attributed to skills, subagents, plugins and MCP servers, with a recommendation for anything above ten per cent. The guide on what uses up the Claude Code 5-hour window goes through each cause with what people have reported.

These are the prices and the usage descriptions on claude.com/pricing on the date at the top of this page. Anthropic describes each plan's allowance as a multiple of another plan's, and does not publish a count.

One line on the Team plan matters if your company pays for it: "Monthly API credits included with Team plans can't be used in Claude Code or the Claude apps, including for extra usage." The seat's allowance is what Claude Code draws on.

Claude Code stops and tells you when the window resets. From Claude Code v2.1.234, it can wait and continue the interrupted task on its own after the reset. The interactive-mode docs say this automatic continue is on by default in interactive sessions signed in with a Claude subscription, stops after three automatic continuations, and does not start on its own for a reset more than 24 hours away. /rate-limit-options lets you choose what to do next.

The other way through is usage credits, which Anthropic called extra usage until recently. The help centre: "Usage credits allow individuals subscribed to paid Claude plans (Pro, Max, and Ultra) to continue using Claude after reaching their included usage limits", and they "are billed at standard API rates". They apply to both Claude conversations and Claude Code. You turn them on under Settings > Usage on claude.ai, where "Adjust limit" sets a monthly spend limit, or run /usage-credits inside Claude Code, which opens that page. Once they are on, /usage shows a usage-credits row with this month's spend against your limit.

Usage credits do nothing until you enable them under Settings > Usage and add funds. With them off, the limit is a hard stop and costs nothing extra. With them on, the stop goes away and the bill begins, which is the trade.

Three places, all Anthropic's. /usage in Claude Code shows your plan usage bars for both windows, your activity stats, and the breakdown of what is behind them. Settings > Usage on claude.ai shows the same two bars and your next reset times. And a custom status line can print the figures on every turn, because Claude Code passes rate_limits.five_hour.used_percentage, rate_limits.seven_day.used_percentage and each window's resets_at to your status line script. The guide on how to check your Claude Code usage walks through each, and the tools that keep a history.

Each member of a Team or Enterprise plan has their own 5-hour and weekly allowance, sized by their seat. The person who runs the plan sees adoption in the analytics dashboard and, once usage credits are on, a spend report of what each member spent past their seat, refreshed daily with a one-day delay. Anthropic's pages do not describe a view of each member's remaining 5-hour or weekly allowance. The guides on what your team spends on Claude Code and who went past their plan cover what an admin can and cannot see.

Anthropic's own screens show where you are now, on one machine. OpenReason Intel, a dashboard for Claude Code and Codex usage, keeps the record of where you were. Its reporter, or-otel, sets itself as Claude Code's status line, reads the two windows' used percentage and reset time that Claude Code passes to it, sends those figures, and then runs your own status line if you had one. For one person, that gives you four things Anthropic's screens do not:

• Your 5-hour and weekly windows over days and weeks, with each time a window filled and when it reset.
• The sessions and turns that were running when a window filled, each with its estimated share of the window, which is what the question "what used it all up?" needs.
• Every machine you use in one record, where /usage counts only the machine it runs on.
• Claude Code and Codex side by side, and a record you can show your lead if your company pays and you need a bigger seat.

It does not warn you. It is a record you look at, not an alarm. What you can do instead: keep the window in view as you work with a status line that prints your 5-hour and weekly percentages (or-otel keeps your own status line running), stop or start a fresh session when it gets high, and look afterwards at which turns filled it. It reports Claude Code in a terminal, and Codex. Claude Desktop has no status line, so sessions there are not reported. The usage leaves your machine and goes to OpenReason; the engineers' page lists what is sent.