You hit the wall mid-task, Claude tells you to come back later, and the obvious question is the one nobody answers cleanly: when exactly does the limit come back? The short answer is that there are two clocks running at the same time, they reset on completely different schedules, and only one of them is tied to anything you did.
This page collects what Anthropic actually publishes about reset timing, quoted verbatim, and marks every place where the company publishes nothing. There are no estimated message counts here and no numbers copied from forum posts. Where a figure does not exist in Anthropic’s own documentation, it says Not published and lists the pages that were checked.
The short answer: two clocks, not one
Every paid consumer plan — Pro, Max 5x and Max 20x — runs two independent usage limits:
- A five-hour session limit. Anthropic’s Pro and Max pages both use the identical sentence: “Your session-based usage limit will reset every five hours.”
- A weekly limit that applies across all models. On the Pro page: “Weekly limits reset at a fixed time each week that is assigned to your account.” On the Max page: “The weekly limit resets at a fixed time each week that is assigned to your account.”
Those two sentences carry the whole practical difference. The five-hour clock is relative and moves with your usage. The weekly clock is absolute and does not move at all — Anthropic states it explicitly: “Your reset day and time stay the same regardless of when you start using Claude or when your subscription begins, and you receive your full weekly allowance each cycle.”
So if you are blocked, the first thing to establish is which limit blocked you. Waiting five hours does nothing for a weekly limit, and upgrading a plan does not move a reset day.
The five-hour session window, in Anthropic’s words
Three separate help pages describe the five-hour window, and they are consistent with each other:
“Your session-based usage limit will reset every five hours.” — What is the Pro plan? and What is the Max plan?
“Your plan’s included usage limit will reset every five hours once you reach it. Usage credits don’t affect this reset timing.” — Manage usage credits for paid Claude plans
“Your session limits reset every five hours as usual.” — Manage usage credits for paid Claude plans
The phrase worth noticing is “once you reach it.” That is the only published statement anywhere about what the five hours are measured from, and it points at your own usage rather than at a clock hour. It is not a schedule shared by all users, and it is not tied to midnight, to your billing date, or to the top of the hour.
What is not published about the session window
- The exact event that starts the session clock. Not published. The pages state the window’s length and that Settings shows time remaining; they do not define the starting event beyond “once you reach it.”
- How many messages a session contains, on any plan. Not published — and by design. The Pro page says the count “will vary based on message length, including the length of files you attach, the length of your current conversation, and the model or feature you use.”
- Any token figure for a session on a subscription plan. Not published.
Pages checked for these three: What is the Pro plan?, What is the Max plan?, How do usage and length limits work?, Usage limit best practices, Manage usage credits for paid Claude plans, Models, usage, and limits in Claude Code, and the public pricing page at claude.com/pricing. None of them contains a per-plan message or token number.
The weekly limit, and the part most people miss
The weekly limit “applies across all models,” in the wording used on both the Pro and Max pages. It resets at a fixed time each week that is assigned to your account — you do not choose it, and it does not shift when your usage patterns shift.
The part most people miss is that the weekly limit is not a single bar. The Usage limit best practices page describes what Settings actually shows:
“Weekly limits: Check when your plan’s weekly usage limit resets for Opus only and all other models.”
That sentence describes two weekly readouts: one for Opus by itself, one for everything else. Practically, that means a week spent leaning on Opus can exhaust the Opus-only allowance while the general allowance is still healthy — which is why switching models with /model sometimes gets you working again immediately and sometimes does nothing at all.
Both plan pages also carry the same open-ended reservation, which is worth reading literally rather than skipping: “In addition, to manage capacity and ensure fair access to all users, we may limit your usage in other ways, such as weekly and monthly caps or model and feature usage, at our discretion.” Monthly caps are named there as a possibility. Their size and trigger conditions are Not published.
How to see your own reset time
Because none of the numbers are published, your own account is the only authoritative source for your own reset times. Both plan pages point to the same place: “You can see your next reset time in Settings > Usage.”
The Usage limit best practices page describes the three readouts you will find there:
| Readout | What Anthropic says it shows |
|---|---|
| Plan usage limits | “shows your progress towards both your session limit and weekly limits” |
| Current session | “How much of your plan’s five-hour session limit you’ve used thus far, plus the amount of time remaining in the session” |
| Weekly limits | “Check when your plan’s weekly usage limit resets for Opus only and all other models” |
This panel is available to “Pro, Max, Team, or seat-based Enterprise” users. Note the second row: the session readout gives you time remaining, not a wall-clock reset time. That is the closest thing to a countdown that Anthropic exposes.
The exact messages you will see
Anthropic’s Troubleshoot Claude error messages page publishes the strings verbatim, which is useful because the wording tells you which state you are in:
| Message | What it means |
|---|---|
| “Approaching 5-hour limit.” | A warning, “when you’re approaching your plan’s limit within a five-hour session.” You are not blocked yet. |
| “5-hour limit reached – resets [time].” | A blocking error. This is the one that carries an actual reset time. |
| “5-hour limit resets [time] – continuing with usage credits.” | Not a block. Shown to “paid Claude users with usage credits enabled in Usage settings.” |
Two other messages get confused with limits and are not limits at all. “Your message will exceed the length limit for this chat” is a length error, not a usage error — it is about the context window of one conversation. And “Due to unexpected capacity constraints, Claude is unable to respond to your message” is neither: Anthropic states plainly that “Capacity constraints are not outages” and that they “will not appear on our status page because they represent normal load management.”
Everything you use draws from the same pool
This single sentence explains most surprise limit hits:
“Note that your usage of all different Claude product surfaces (claude.ai, Claude Code, Claude Desktop) counts towards the same usage limit.” — How do usage and length limits work?
An afternoon of browser chat can therefore end a terminal session, and vice versa. The Max plan page lists Claude Code and Cowork as included in “one unified subscription” — unified access and a unified limit are the same coin. If you are trying to work out which tier buys you enough headroom for that combined usage, we walk through the tier maths in Claude Max 5x vs 20x, and the Pro-level baseline in Claude Pro usage limits.
What burns through a window fastest
Anthropic publishes a list of factors rather than a formula. From Usage limit best practices, the factors that affect your usage limits are: message length; file attachment size; current conversation length; tool usage (e.g., Research, web search); model choice; effort level; and artifact creation and usage.
How do usage and length limits work? adds the same idea in different words — usage “is affected by several factors, including 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” — plus one specific and easily missed consequence: “Longer conversations that trigger automatic context management consume more of your usage limit.” The advice attached to it is blunt: “Try starting a new conversation if you’re approaching your usage limit in a longer chat.”
Caching works in your favour, and it is documented
Three published statements, all from Usage limit best practices:
- “Content in projects is cached and doesn’t count against your limits when reused.”
- “Similar prompts you use frequently are partially cached.”
- “Every time you reference that content, only new/uncached portions count against your limits.”
The operational takeaway is that re-uploading the same reference document into fresh chats is one of the few genuinely wasteful habits with a documented fix: put it in a project once. The same page also names tools and connectors as “token-intensive” and suggests disabling web search, Research and MCP connectors when a conversation does not need them. No percentage is attached to any of this — the size of the caching benefit is Not published.
Claude Code: same limit, different levers
The Models, usage, and limits in Claude Code page is the most mechanical description Anthropic publishes of where usage goes. Every turn sends three things to the model: “The conversation so far — every previous message in this session,” “Project context — your CLAUDE.md and any files Claude has read,” and “Your new prompt.” The page then names the culprit: “Of these, the first item grows the fastest.”
That single fact is why /clear is the strongest lever available. Anthropic calls it “the single most effective lever for both quality and cost,” with the useful test attached: “if your next prompt would make perfect sense in a brand-new terminal, clear before sending it.” /compact is the mid-task alternative, summarising rather than wiping. One warning is published and worth repeating: “/clear cannot be undone.”
Model choice matters here too, in plain language: “Opus costs several times more per turn than Sonnet, and Sonnet more than Haiku. Spending Opus on routine work is the fastest way to drain a daily limit.” The documented pattern is to “plan with Opus, execute with Sonnet,” which is built in as /model opusplan.
Two Claude Code details bear directly on resets. First, the page distinguishes a full context window from a usage limit explicitly — “Context window full (which is different from a usage limit)” — so a context warning is not something a reset will fix. Second, for Enterprise-seat sign-ins it describes the allowance as “reset on a rolling window,” which is the only place the word rolling appears in this documentation set. API-key sign-ins behave differently again: “No hard stop; the account is charged for what it uses.” If you are still working out which of these applies to you, whether Claude Code is free covers the entry paths, and Claude Max versus Codex usage limits compares this metering model against OpenAI’s.
Not waiting: usage credits
Usage credits are the documented way to keep working through a session limit rather than waiting it out. They are available to “individuals subscribed to paid Claude plans (Pro, Max 5x, and Max 20x),” and the mechanism is described as: “Instead of being blocked when you hit your session limits, you can switch to consumption-based pricing at standard API rates and continue your work without interruption.”
Details that are published, verbatim:
- Credits do not change reset timing: “Usage credits don’t affect this reset timing,” and “Your session limits reset every five hours as usual.”
- They are prepaid: “You’ll then need to prepay to cover usage beyond your plan limits,” with optional auto-reload and a monthly spending cap.
- “There is a daily redemption limit of $2000.”
- They cover the terminal too: “Usage credits apply to both Claude conversations and Claude Code terminal usage.”
- Expiry: “In most cases, usage credits do not expire. However, in certain jurisdictions such as Japan, usage credits expire six months after purchase starting September 10, 2026.”
- Mobile subscribers must enable them on the web: credits “are only available if you enable and purchase them on the Claude web app.”
The credit rate is “standard API rates” — pointed at the pricing page rather than restated in the help centre. Anthropic does not publish a conversion between a subscription session and a dollar amount of credits: that is Not published.
Does the reset behaviour differ by plan?
No. The reset mechanics are identical across Pro, Max 5x and Max 20x — the same five-hour session window, the same account-assigned weekly reset. What differs is how much fits inside each window.
| Plan | Published price | Published capacity statement | Reset behaviour |
|---|---|---|---|
| Pro | “$20 per month (US)”; claude.com/pricing shows $17/month with the annual plan, billed $200 up front | “At least five times the usage per session compared to our free service” | 5-hour session + account-assigned weekly |
| Max 5x | “$100 per month” | “five times more usage per session than the Pro plan” | Identical |
| Max 20x | “$200 per month” | “20 times more usage per session than the Pro plan” | Identical |
Anthropic attaches two caveats to the Max prices that are easy to miss: “These prices are for web subscriptions only. Mobile pricing may vary depending on your app platform,” and “Price and plans are subject to change at Anthropic’s discretion.”
Frequently asked questions
When does my Claude usage limit reset?
If you hit a session limit, five hours — “Your session-based usage limit will reset every five hours.” If you hit a weekly limit, at “a fixed time each week that is assigned to your account,” visible in Settings > Usage. The blocking error itself names the time: “5-hour limit reached – resets [time].”
Does the weekly limit reset on the same day every week?
Yes, per Anthropic: “Your reset day and time stay the same regardless of when you start using Claude or when your subscription begins, and you receive your full weekly allowance each cycle.” Which day you were assigned is not published — check Settings > Usage.
Can I make my limit reset sooner?
Nothing in Anthropic’s documentation shortens a window. The published alternatives are to wait, upgrade, or enable usage credits — and credits explicitly “don’t affect this reset timing.” Switching to a lighter model with /model may let you keep working if what you exhausted was the Opus-only weekly allowance.
Does Claude Code have its own separate limit?
Not on a consumer subscription. Usage across “claude.ai, Claude Code, Claude Desktop” counts toward the same limit. Enterprise seats and API keys are metered differently — a pooled allowance “reset on a rolling window” for seats, and “No hard stop” for API keys.
Why did I hit a limit without sending many messages?
Because messages are not the unit. The published factors are message length, attachment size, current conversation length, tool usage, model choice, effort level, and artifact creation. A single long conversation with Research enabled on Opus is not comparable to a short exchange on Haiku, and Anthropic never claims otherwise.
Sources
Every quotation above was read directly from these pages. Last verified 14 August 2026.
- What is the Pro plan? — Claude Help Center
- What is the Max plan? — Claude Help Center
- How do usage and length limits work? — Claude Help Center
- Usage limit best practices — Claude Help Center
- Troubleshoot Claude error messages — Claude Help Center
- Manage usage credits for paid Claude plans — Claude Help Center
- Models, usage, and limits in Claude Code — Claude Help Center
- Plans & Pricing — claude.com
Anthropic changes plan mechanics without notice, and these help pages carry no visible last-updated dates. Treat the quotations here as a snapshot of 14 August 2026 and check Settings > Usage for anything that concerns your own account.
If waiting for a reset does not help because the conversation itself is full, you are looking at the wrong ceiling. See the Claude Code context window for the sizes, the auto-compact threshold, and what caching costs.
The equivalent question on the OpenAI side is answered in Codex usage limits by tier, where the five-hour allowances are published per model.
Update, 22 September 2026: Anthropic now documents a weekly limit whose reset time is assigned to your account, and publishes no number for either clock
This page is about when your limit resets. As of this week, the documentation it describes has changed in a way that makes the question harder to answer rather than easier, and the change is worth spelling out precisely because it is easy to read the new wording as reassurance.
Both the Pro and the Max help-centre articles now carry an “Updated this week” stamp, and both have gained a paragraph that was not there when we last read them on 14 September 2026. The substance, in the vendor’s own terms: the plan also has a weekly usage limit that applies across all models; that weekly limit resets at a fixed time each week that is assigned to your account; and you receive your full weekly allowance each cycle.
Read that as three separate statements, because they do three different things.
1. There are now two clocks, and they are documented differently
The five-hour session window that most of this page deals with is a rolling clock: it starts when you start, and it is legible because you can watch it start. The weekly limit is a fixed-phase clock: it has a specific moment each week at which it resets. Those are not variations on a theme. A rolling window can be reasoned about locally — you know when you began. A fixed-phase window can only be reasoned about if you know the phase.
And the phase is the part that is not published. The wording is that the reset time is assigned to your account — not that it is Monday, not that it is midnight UTC, not that it is the anniversary of your signup. It is a per-account value, and the documentation does not tell you how to look yours up. The single most operationally useful fact about a weekly limit is its reset moment, and it is the fact the page declines to give you.
The practical consequence is that a very common piece of advice — some version of “your weekly allowance comes back on Monday, so push the big refactor to Monday morning” — is structurally wrong, not merely imprecise. If reset phases are distributed across accounts, then the correct advice for any individual is unknowable from documentation and can only be established by observation: note the moment the ceiling lifts, once, and you have your account’s phase for good. That is a five-second measurement that no amount of reading will substitute for.
2. “Your full weekly allowance each cycle” answers a question nobody was asking
The sentence confirms that the weekly bucket does not carry over and does not accumulate — each cycle starts at full, regardless of what you did last cycle. That is genuinely useful, and it is also the only one of the three statements that carries no ambiguity.
What it does not do is tell you how large the allowance is. No number appears anywhere in the new paragraph, for either plan, in any unit. There is no token figure, no message figure, no hour figure, and no statement of what “across all models” means for a user who mixes a cheap model and an expensive one. A ceiling whose height is undisclosed is not a limit you can plan against; it is a limit you can only discover by hitting.
3. The bigger change: hitting the limit is no longer a wall, it is a price change
A separate help-centre article, dated 10 August 2026 and not previously recorded here, documents usage credits — and it changes what “reaching your limit” even means. The published mechanics: credits are prepaid; once your subscription allowance is exhausted, further usage is billed at standard API pricing rates; you can set a monthly cap or leave it unlimited; automatic reload is available; there is a daily redemption limit of $2,000; and Claude Code and claude.ai usage are metered together rather than separately.
Put the two documents side by side and a specific, uncomfortable thing falls out. Before credits, hitting a limit produced a stop — annoying, but free and unambiguous. With credits enabled, the same moment produces a discontinuity in your marginal cost: the price of your next token jumps from “already paid for” to standard API rates, instantly, with no intermediate state. And because the weekly reset phase is undisclosed and the weekly ceiling size is undisclosed, you cannot know in advance when that jump will happen. You have an unpriced boundary at an unknown time, with automatic top-ups available to carry you across it without interrupting anything.
We are not suggesting anything is hidden — every one of these facts is published, and the monthly cap exists precisely to bound the exposure. The point is narrower and, we think, more useful: these are three separate documents whose interaction is not documented anywhere, and the interaction is the thing that determines your bill. If you turn on usage credits, set the monthly cap at the number you would be unhappy to see on a statement, not at unlimited, and treat auto-reload as a convenience that removes the one signal telling you the boundary was crossed.
The “metered together” clause deserves its own sentence. If Claude Code and claude.ai draw from the same pool, then an afternoon of chat usage moves the moment your coding agent starts costing API rates, and vice versa. Anyone budgeting the coding side in isolation is budgeting one input to a two-input function.
A quiet removal: the annual price is no longer on the page
One more change in the same update, easy to miss and worth recording. Pro is still documented at $20 per month (US), and Max is still documented at $100 for the 5x tier and $200 for the 20x tier, both described as monthly subscription only. But the annual option for Pro is now described as “billed once yearly at a discounted rate” — the dollar figure that was previously on the page is no longer there. The annual price is now officially not published on that page.
That is a small disclosure regression with a concrete effect: the page no longer supports an annual-versus-monthly calculation. You cannot compute the break-even without the number, and we are not going to supply one from memory or from a third-party summary. Where a figure has been withdrawn, we record the withdrawal rather than filling the hole — a number we cannot source today is worth less than an honest blank.
There is also a regional term: in Japan and certain other markets, credits purchased from 10 September 2026 carry a six-month expiry. Prepaid balance that expires is a materially different product from prepaid balance that does not, and if you are in an affected market that clause should govern how much you pre-buy at a time.
What this changes about the advice on this page
Most of what is written above still holds, because it concerns the rolling five-hour window and that mechanism has not changed. What the weekly clock adds is a second failure mode that the five-hour advice cannot address:
- Measure your own reset phase once, then write it down. It is account-specific and undocumented, so it cannot be looked up, but it also does not change. One observation buys you the answer permanently.
- Stop treating the five-hour window as the constraint if you are running out weekly. The two ceilings fail differently: a five-hour exhaustion is a pause, a weekly exhaustion is either a week-long pause or, with credits enabled, a switch to API pricing. Spacing your sessions fixes the first and does nothing for the second.
- Decide the credits question before you hit the boundary, not at it. The moment you are blocked is the worst moment to evaluate whether unlimited auto-reload is a good idea.
- If you are hitting the weekly ceiling consistently, the plan tier is the lever, not your scheduling. Our comparison of Max 5x against 20x is the relevant page for that decision, and the honest framing is that you are buying a higher undisclosed ceiling, which is a strange thing to buy but is nonetheless the thing on offer.
For the five-hour mechanics and the day-to-day pacing that follows from them, our guide to Claude Pro usage limits remains the more detailed treatment. And if the weekly ceiling is pushing you to price out alternatives, note that the subscription products from other vendors have adopted exactly the same two-clock structure — a short burst window inside a much tighter rolling weekly cap — which we work through with published numbers in our comparison of the GLM Coding Plan against Claude Code. The difference worth knowing is that some of those vendors publish the size of the weekly bucket.
Qualifiers
- Every figure here is from the vendor’s own help-centre articles, read on 22 September 2026. The weekly limit paragraph was not present when the same pages were read on 14 September 2026; we are reporting the diff between two of our own readings, which dates the documentation change and not the policy change. When the underlying limit actually took effect is officially not published.
- No weekly allowance figure exists to report. Any specific token or message number you encounter for the weekly cap did not come from these pages. We have not adopted third-party estimates and we would treat them with suspicion.
- The reset phase is account-specific, so any reset time stated here or anywhere else as a general fact is wrong by construction, including for us.
- The $2,000 daily redemption limit applies to credit redemption as documented; it is not a statement about what a subscription can spend.
- The six-month credit expiry is regional. Do not generalise it, and do not assume your market is exempt without checking.