Claude usage limit reached: which limit did you hit?

"Usage limit reached" means one specific window is spent, and which window it is decides whether you are waiting hours or days. A Claude subscription runs three separate ceilings that can produce this message: the rolling five-hour session window, the weekly caps, and a model-specific cap you can often step around in seconds. This page is about the hard stop. If you are still only seeing a heads-up, the approaching weekly limit warning is the page you want.

Which limit did you hit?

The wording of the message is the fastest diagnosis you have. Claude Code documents four of them, and they are not interchangeable:

• "You've hit your session limit" means the rolling five-hour session window is spent. That window opened with your first message and runs about five hours from there, so the wait is measured in hours.
• "You've hit your weekly limit" means the weekly cap across all models is spent. Weekly caps reset at a fixed time each week assigned to your account, so this wait can be anything from hours to several days.
• "You've hit your Opus limit" means the separate weekly ceiling for Opus is spent while your general capacity is very likely still fine.
• "You've hit your Sonnet limit" is the same idea applied to that model family.

The distinction that matters most in the moment: the session and weekly limits apply across all models, so switching with /model does nothing for them. The model-specific limits do not, so switching to a model outside that family gets you working immediately. That single fact is the difference between a five-second fix and an afternoon of waiting, and it is covered in more depth in the Claude Opus usage limit.

The phrasing is friendlier in the web and desktop apps, but the clocks are identical: claude.ai, Claude Desktop, and Claude Code all draw from the same pool, so a heavy terminal morning and a long browser conversation compound.

Read it off your usage screen instead of guessing

If the message scrolled past or you are not sure which one you saw, your account tells you directly. Open Settings > Usage in the Claude web or desktop app, or run /usage in Claude Code. Both show one row per window with a progress bar, a percentage from 0 to 100, and the exact time that window next resets, displayed in your local timezone.

What to look for:

• Session bar full, weekly bars healthy. You burned a short window hard. This is the good outcome; you are counting hours.
• Weekly bar full. Check the reset time before you rearrange anything. Weekly resets are personal to your account rather than a shared calendar hour, which is why no article can quote you one. See when Claude limits reset.
• Opus bar full, general weekly bar fine. Switch models and carry on. Nothing to wait for.

/usage in Claude Code adds a breakdown of which skills, subagents, plugins, and MCP servers ate the most recently. Those attribution figures come from local session history on that machine, so they miss work you did elsewhere; the plan usage bars themselves cover your whole account. Full walkthrough: how to check your Claude usage.

Is "Claude is at capacity" the same thing?

No, and this one trips up a lot of people. A capacity notice is about Anthropic's servers, not your account. Claude Code surfaces it as a 529 Overloaded error with wording along the lines of "The API is at capacity, this is usually temporary. Try again in a moment." Anthropic's own documentation is explicit that this is not a usage limit and does not count against your quota.

Three practical differences: it clears on its own, usually in minutes, rather than at a scheduled reset; capacity is tracked per model, so switching with /model often gets you moving again; and status.claude.com will show it, whereas a usage limit never appears on a status page.

A third family is worth separating too: "Server is temporarily limiting requests" and a bare 429 are throttling, meaning requests arriving too fast rather than an exhausted allowance. Claude Code retries temporary 429s automatically and only surfaces the error once retries run out. If you build against the API as well, the Claude rate limit error untangles the set.

What you can do right now

In rough order of how often it helps:

• Switch models, if the message named one. /model in Claude Code, or the model picker in the app. Sonnet handles the large majority of coding work well and Haiku is faster still for mechanical tasks.
• Let Claude Code wait for you. On recent versions, when a limit stops Claude mid-task the session stays open and shows a line like Usage limit reached · continuing automatically at 3:45pm · esc to cancel, then picks the task up on its own after the reset. Leave the session open and it resumes without you; /rate-limit-options controls the behavior.
• Spend the pause on the half of the work that needs no model. Review the last diff properly. Write the test you were going to ask for. Decide what you actually want next.
• Restructure the remaining work into smaller, sharper asks. Vague requests like "improve this codebase" trigger broad scanning; naming the file and the change does not. Anthropic's guidance is to plan the conversation first, batch related requests into one message, and be specific rather than exploratory.
• Clear between unrelated tasks. Claude Code resends the whole conversation with every request, so a session left open all day keeps drawing usage for its entire history. /clear costs nothing.

Does upgrading or buying extra usage help?

Sometimes, and it depends on which clock you hit. There are exactly three things that change your situation with money, and one that changes it for free.

• Upgrade the tier. This is the only change that lifts the ceilings themselves. As of September 2026, Claude Pro is $20 a month billed monthly or $17 billed annually, and Claude Max comes in two monthly tiers, Max 5x at $100 and Max 20x at $200, described by Anthropic as five times and twenty times more usage than Pro. Prices change, so confirm at claude.com/pricing. Whether the jump is worth it is a question of pattern, not budget: see Claude Max vs Pro.
• Turn on usage credits. Pro, Max 5x, and Max 20x subscribers can opt in to usage credits and keep working past the included allowance. They are off unless you enable them, in Settings > Usage or via /usage-credits in Claude Code, and they bill at standard API pricing rates rather than expanding your plan. You can set a monthly spend cap and turn them off at any time. One detail worth knowing first: the prompt cache lifetime is an hour on subscription usage and drops to five minutes once you are drawing on credits, so long sessions get noticeably more expensive.
• Move the heaviest work to metered API billing. A different product with a different bill, not a bigger allowance. The tradeoff is laid out in Claude Pro vs API cost.
• Wait. The honest default. Both clocks refill on their own, nothing is lost, and there is no penalty on your account for having hit the limit.

What will not help: there is no setting that raises a cap, no support ticket that grants a one-time increase, and no trick involving new chats or a fresh install. The limit belongs to the account, not the conversation.

If you almost never see this message

Worth flipping the question, because most subscribers land on this page once out of curiosity and then never hit a wall again. Both clocks are ceilings, not balances. Every five-hour window and every weekly cycle refills to the same level whether you used all of it or none of it, so capacity you never reach is paid for and then quietly expires.

Tokens for Good exists to catch that recurring surplus. Your Claude Code claims a queued nonprofit, researches its real-world impact against a fixed methodology with citations, and submits a structured report. Every organization is researched twice by independent contributors, then validated, consolidated, scored deterministically, and human-reviewed before it reaches the public directory. It runs on the subscription you already pay for with no separate API cost, and it can run on a schedule in the background, so it fills windows you were never going to touch. To be clear, "tokens" here means AI model tokens, not crypto: no coin, no wallet, no blockchain. See how the research works or the docs, which take about a minute, or start with what Tokens for Good is.

Frequently asked questions