What Are Claude Code Usage Limits?

What Are Claude Code Usage Limits?

Search this question and you will find tables. Prompts per five hours by plan. Hours of Sonnet per week. Token quotas to the nearest thousand.

Almost none of those numbers come from Anthropic. The company documents the limit machinery in considerable detail and declines, in writing, to publish how big any of it is.

That absence is the single most useful fact about Claude Code usage limits, and it should change how you plan around them.

What Does Anthropic Actually Publish?

Mechanisms, not amounts.

There are three limits that can stop a Claude Code session on a subscription plan: a rolling five-hour session limit, a weekly limit, and per-model limits that apply separately to the Opus and Sonnet families.

Context limits exist too, but they are a different system and get confused with usage limits constantly.

What Anthropic does not publish is a number for any of them, on any plan.

Its pricing FAQ addresses the question head-on and refuses the figure: how far your allowance goes depends on how long and complex your conversations are, which model you choose and which features you use, so no fixed message count exists.

Anthropic also reserves the right to move the goalposts, stating on several pages that it may adjust usage limits at its discretion, including through weekly or monthly caps and limits on particular models or features.

Is your content strategy running on guesses too? Get a free SEO audit at seo-stuff.com/free-audit and work from real numbers.

How Do the Three Limits Work?

The five-hour session window

The five-hour limit governs how much you can use Claude within a five-hour period. The window opens when you send your first message and closes five hours later, rather than resetting at a fixed hour.

Hit it and Claude Code tells you when it lifts. The error reads You've hit your session limit · resets 3:45pm.

One plan is exempt: Anthropic's top consumer tier, now called Ultra, has no five-hour limit for verified accounts.

The weekly limit

Paid plans add a weekly limit on top, applying across all models.

It resets at a fixed time each week assigned to your account, not on a rolling basis and not at a time you choose. The error is You've hit your weekly limit · resets Mon 12:00am.

The interaction between the two windows is the part that catches people, and the Claude Code documentation states it plainly: "Usage counts against the session and weekly allowances at the same time. A single burst of heavy activity, such as a large workflow fanout, can exhaust the weekly allowance before the session window resets."

So a heavy afternoon does not just cost you the rest of that session. It can cost you the rest of the week.

The per-model limits hardly anyone documents

This is the best-hidden mechanism in the set, and it appears only in the Claude Code developer documentation. Not on a single plan page.

Alongside the session and weekly errors there are two more: You've hit your Opus limit · resets 3:45pm and You've hit your Sonnet limit · resets 3:45pm.

The documentation explains the difference that matters: "The session and weekly limits are shared across all models, so switching models doesn't restore access. The Opus and Sonnet limits each apply only to requests to that model family, so switching to a model outside the family with /model keeps you working."

That is a genuinely actionable distinction. On a session or weekly limit, switching models achieves nothing. On a model-family limit, /model gets you back to work immediately.

How big are the Opus and Sonnet sub-limits? Anthropic does not say, anywhere, in any form.

Context limits are not usage limits

A great deal of reported "hitting the limit" is actually the context window. Anthropic treats these separately: usage limits are how much you can do over time, length limits are how much fits in one conversation.

Claude Code is explicit that its own context warning is a different animal, describing a context or auto-compact warning as not a usage limit, raised when a conversation approaches the threshold where older history gets summarised to free space.

How Do You Check Your Own Usage?

The command is /usage. Per the documentation it will "Show session cost, plan usage limits, and activity stats", and on a Pro, Max, Team or Enterprise plan it "includes a breakdown of what counts against your plan limits". /cost and /stats are aliases for the same screen.

Two things on that screen are worth knowing about. The breakdown attributes recent usage to skills, subagents, plugins and individual MCP servers as percentages, and a recommendations list proposes specific changes with the share of usage each would recover, so a bloated project memory file shows up as a line item rather than a mystery.

The caveat is significant and easy to miss: "The figures are approximate and computed from local session history on this machine, so usage from other devices or claude.ai is not included." If you use Claude on a phone, in a browser and on two laptops, /usage is a partial view by design.

The authoritative figures live in the Claude Console and in your account's usage settings.

Two Anthropic pages disagree about which command to use

An Anthropic support article on using Claude Code with a Pro, Max or Ultra plan tells you to monitor your remaining allocation with the /status command.

The Claude Code documentation says /status opens the settings interface on the status tab, showing version, model, account and connectivity. No allocation. The command that shows plan usage limits is /usage.

Want a clear read on your organic performance? Claim your free audit at seo-stuff.com/free-audit.

What the Published Multiples Do Tell You

Anthropic publishes no absolute figures, but it does publish relative ones, and they are consistent enough to plan with.

Pro is the unit of account at one times itself. Max is documented as five times the Pro plan's usage, both every five hours and every week. Ultra is published as ten times Pro's weekly usage, with no five-hour limit for verified accounts.

The plan previously sold as Max 20x is now called Ultra, existing subscribers moved across with no price change, and the published multiple for the same two hundred dollars a month is ten times Pro weekly.

Anthropic nowhere reconciles twenty with ten, or explains how removing the five-hour cap trades against the weekly figure. Any article still quoting "20x Pro usage" is describing a plan that no longer exists.

Team plans carry the most precise numbers Anthropic publishes anywhere, oddly enough on a plan page rather than the pricing page: Standard seats at 1.25 times Pro's per-session allowance and Premium seats at 6.25 times, applied per member rather than pooled across the team. Several third-party guides claim Anthropic does not publish these. It does.

Enterprise has no published ceiling at all. On usage-based Enterprise there are no per-seat usage limits, and an admin sets a dollar amount or unlimited.

Guessing at your own numbers is no better. Get a free SEO audit at seo-stuff.com/free-audit and see where your traffic actually comes from.

What Happens When You Hit a Limit?

It blocks. It does not quietly downgrade you to a smaller model, which is a common assumption. In Anthropic's words, "Claude Code blocks further requests until the reset time shown in the message."

Three things soften that. There is a small wrap-up allowance that lets Claude finish a response in progress when a five-hour limit lands mid-answer, capped and counted against your weekly limit, and on Pro it applies at most once per weekly period.

Usage credits can be enabled on Pro, Max and Ultra and are billed at standard API rates, with /usage-credits handling the setup or sending a request to your admin on Team and Enterprise. And /rate-limit-options lists the ways forward when a limit blocks a request.

The most useful behaviour is the one least written about. Claude Code can wait out the limit and resume by itself: "When a claude.ai usage limit stops Claude mid-task, Claude Code waits in the open session and continues the task on its own after the limit resets. Automatic continue is on by default in interactive sessions signed in with a claude.ai subscription."

It is not unconditional. The wait is not offered automatically when a reset is more than twenty-four hours away, which a weekly limit often is, and Claude Code re-arms the wait at most twice in a row before stopping.

It also does not resend your last message, prompting Claude to pick the task up instead.

Limitations and What Anthropic Leaves Unanswered

Four gaps are worth knowing about before you build a workflow on assumptions.

Rollover is undefined. No Anthropic page states that unused session or weekly capacity carries over, and none states that it does not. The nearest statement is that you receive your full weekly allowance each cycle. Purchased usage credits are a separate matter and do carry expiry rules.

The support article whose title promises this exact topic largely does not deliver it. Anthropic's "Models, usage, and limits in Claude Code" article never mentions the five-hour window or the weekly limit, and its table explaining how usage is metered has exactly two rows, for an Enterprise seat and an API key.

Pro, Max, Ultra and Team, which is most Claude Code users, are absent from a table whose job is to tell you how your usage is counted.

Promotional changes are frequent and the baseline is invisible. A promotion raised Claude Code weekly limits by half for four months in 2026, and when it ended the uplift was partly kept, leaving weekly limits a quarter higher than before.

That percentage multiplies a number Anthropic has never published, so it cannot be turned into anything absolute.

One mechanism was announced and never documented. Anthropic's own news post describes removing a peak-hours limit reduction on Claude Code for Pro and Max, which means limits were once cut at busy times.

No help article covers it, and nothing first-party says whether it still applies to Team or Enterprise.

How Do Subscription Limits Compare to API Rate Limits?

API rate limits are published in full, as requests per minute and input and output tokens per minute, by usage tier, refilling continuously rather than resetting at intervals.

If you found a precise table of Claude limits, there is a good chance you found this one, and it does not apply to a Claude Code subscription.

The practical difference is that on an API key or a cloud provider there is nothing to wait for: usage is metered per request, so there is no reset.

One trap to avoid is that the monthly API credits bundled with some plans cannot be spent in Claude Code or the Claude apps, which is a different pot from the usage credits described above.

Your analytics should answer questions, not raise them. Start with a free audit at seo-stuff.com/free-audit.

Final Thoughts

Treat every hard number you read about Claude Code usage limits as unsourced until you find it on an Anthropic page, because most of them are not there.

What is documented, and has been stable, is the shape: a rolling five-hour window, a weekly window that bills simultaneously, separate Opus and Sonnet ceilings you can route around with /model, and a /usage screen that tells you which of your own skills, plugins and MCP servers is spending the allowance.

Plan on the mechanisms and let /usage be your ground truth. It is approximate, local to one machine, and still more reliable than any table on the internet.

If you are weighing which plan to pay for rather than how to live inside one, our guide to Claude AI pricing covers how the tiers compare.