When you see "Claude usage limit reached"
As of 2026-11-09, when Claude Code shows "You've hit your session limit", "You've hit your weekly limit", "You've hit your Opus limit" or a message starting "Fable limit reached", read the reset time in the message, then wait for the reset, switch to another model (this only helps with the Opus, Sonnet and Fable limits), turn on usage credits with /usage-credits, or upgrade your plan. The session limit is the 5-hour allowance and the weekly limit recovers at your account's weekly reset time; Max and Ultra plans also have a separate weekly Fable limit. On 2026-10-30 Anthropic renamed Max 5x to Max and Max 20x to Ultra; prices did not change. Verified Ultra accounts have no 5-hour limit, so on a verified Ultra account the limit you hit is the weekly limit or the Fable weekly limit (Help Center and Claude Code error reference fetched 2026-10-26, rechecked 2026-11-06 and 2026-11-09). You can also keep working on an API billed per token.
Updated 2026-11-10
One key for Claude, GPT and leading Chinese models, billed per token, up and running in minutes.
- Billed per token — live rates on /models
- Pay by card (Visa / Mastercard), Apple Pay, Google Pay or crypto
- Self-service: top up and activate instantly
Four things to know
Rolling window
A short-cycle cap on a rolling window. It restores on its own — nobody needs to be contacted.
Weekly cap
A longer-cycle total. Once you hit it, waiting out the 5-hour window will not help.
Opus, Sonnet and Fable limits
As of 2026-11-09, an Opus or Sonnet limit covers that model family only, so switching with /model to a model outside it keeps you working. Max and Ultra can spend up to 50% of their weekly limits on Fable, then continue on usage credits or switch models. The +50% weekly boost ran from 2026-05-13 to 2026-09-13 (see the timeline).
Two different reset times
On the QCode side: the daily limit resets at 00:00 Beijing time; booster quota resets separately at 02:00 the next day.
What the message actually says
As of 2026-11-09 the Claude Code error reference lists these plan-limit messages: "You've hit your session limit · resets 3:45pm", "You've hit your weekly limit · resets Mon 12:00am", "You've hit your Opus limit · resets 3:45pm" and "You've hit your Sonnet limit · resets 3:45pm" (fetched 2026-10-26, rechecked 2026-11-09).
While Claude Code waits for a reset the session shows "Usage limit reached · limit resets 3:45pm". Claude Code blocks further requests until the reset time shown; the session and weekly limits are shared across all models, so switching models doesn't restore access - only a per-family limit lets you keep working by moving outside that family.
That time is the tell: a few hours out generally means the rolling 5-hour window; a few days out means the weekly cap. Both are set by Anthropic per subscription tier and have nothing to do with your network, client version or API key. Max and Ultra plans also have a separate weekly limit for Fable, and that message starts with Fable limit reached.
If you reach the 5-hour session limit while Claude Code is partway through a response, it may keep working briefly to finish the step in progress, and you will see Usage limit reached · wrapping up. This wrap-up allowance is small and capped, and the usage counts toward your weekly limit; it applies only to a response already in progress, not to a new message sent after the limit is reached.
On Pro it applies up to once per weekly usage period; on Max it can apply each time you reach a 5-hour session limit, within your weekly limit. Verified Ultra accounts do not have a 5-hour limit, so the wrap-up allowance does not apply; if an Ultra account still has a temporary limit, the wrap-up allowance works the same way as on Max until that limit is removed.
Anthropic says the feature is rolling out gradually, so it may not be on your account yet (Help Center article 11145838, fetched 2026-11-06). If you have usage credits on and see "You've hit your monthly spend limit", that is your own spend cap: raise it in Settings > Usage. "Server is temporarily limiting requests (not your usage limit)" is a short server-side throttle that does not touch your quota: wait briefly and try again (Claude Code error reference, rechecked 2026-11-09).
The current official wording and the newest entry
The Help Center article 15910845, fetched 2026-09-20, reads in the past tense: the +50% boost "ran from May 13 through September 13, 2026". Same article: starting September 14, 2026, weekly limits in Claude Code are 25% higher than before the promotion, for Pro, Max, Team and legacy seat-based Enterprise; 5-hour limits were outside this promotion.
Rechecked 2026-10-08: the Help Center article is unchanged; the latest Help Center release-notes entry was dated 2026-09-28 (Claude Sonnet 5.5 launch), and this round's fetch on 2026-10-20 shows the newest entry dated 2026-10-07 (Claude Haiku 5.5 launch, Max and Team plans gaining monthly API credits). Nothing new on weekly limits since.
Timeline
The timed weekly-limit promotion recorded in the Help Center begins; the same note says 5-hour limits were not affected by it. From early August 2026, many developers publicly log "most of my allowance gone in minutes", usually alongside parallel threads and whole-repo sweeps.
On 2026-08-18 Help Center extended the +50% weekly boost to 2026-08-31 and mentioned capacity pressure; that was the wording then. The boost finally ran out on 2026-09-13: Anthropic says it had planned to return weekly limits to their original levels, and instead kept part of the boost from 2026-09-14.
The +50% boost is over. Help Center, fetched 2026-09-20, is past tense: the promo ran 2026-05-13 to 09-13, and weekly limits are 25% above pre-promotion from 09-14. From this day part of the boost stays permanently in Claude Code's standard weekly limits, unchanged on the 2026-10-20 recheck.
Work out which one you hit first
What is confirmed
As of 2026-11-09 the Claude Code limit messages mean this (Claude Code error reference and Help Center, fetched 2026-11-09). 1) "You've hit your session limit · resets 3:45pm": the rolling 5-hour allowance is used up, shared by all models; it returns at the reset time in the message. Wait, run /usage-credits, or upgrade; verified Ultra accounts have no 5-hour limit. 2) "You've hit your weekly limit · resets Mon 12:00am": the weekly allowance for all models is used up; weekly limits reset once a week at a fixed time assigned to your account (see /usage or Settings > Usage).
Wait, run /usage-credits, or upgrade; a new plan's limits apply right away. 3) "You've hit your Opus limit · resets 3:45pm" (or "You've hit your Sonnet limit · resets 3:45pm"): a limit for that model family only. Run /model and pick a model outside the family to keep working; the next request re-reads the whole conversation with no cache hits.
The error reference gives no cycle length, so go by the reset time in the message. 4) Fable limit (Max and Ultra): Max and Ultra can use up to 50% of their weekly limits on Fable, drawn from the same weekly limits. Past it, in an interactive session Claude Code asks whether to continue on Fable using usage credits or switch to your default model; if nobody answers that prompt (Remote Control, background or agent-team sessions), the turn ends with "Fable limit reached · continuing on Fable 5.1 uses usage credits, and the prompt to confirm went unanswered — nothing was sent". On Pro, Fable runs on usage credits from the start.
The +50% weekly boost ran until 2026-09-13; its dates are in the timeline above. The Help Center now says it plainly: paid plans have two types of usage limits, a session limit that resets every five hours and weekly limits that reset once a week, and usage across claude.ai, Claude Code and Claude Desktop counts towards the same limit.
The Claude Code error reference adds that one turn is charged against both allowances at the same time. Rechecked 2026-10-26 and again 2026-11-06: usage credits are billed at standard API rates and a monthly spend limit can be set in Settings > Usage; monthly API credits included with Max and Ultra cannot be used for usage credits.
What the official pages leave unstated
Some claim that under load Anthropic silently routes requests to a smaller or quantised model; there is no official statement about it and no first-person evidence at scale in public threads. Being refused, truncated or blocked by a classifier is a different phenomenon from "they swapped the model". Anthropic does not publish the numeric size of each plan's session and weekly limits: the Help Center says the number of messages varies with message length, attachments, the length of the current conversation and the model you use, and it keeps the option of further limits such as weekly or monthly caps or limits by model and feature.
The weekly limit does reset once a week at a fixed time assigned to your account, and there is no single table for every account; your next reset time appears in /usage and Settings > Usage.
The two caps
Rolling 5-hour window
A short-cycle ceiling. It restores automatically at the time printed in the error. You normally hit it through dense use in a short span.
Weekly cap
A longer-cycle total. Once you hit it, waiting for the 5-hour window does nothing. The options Anthropic lists now are waiting for the reset, upgrading your plan (its limits apply right away), or turning on usage credits and continuing at standard API rates. With usage credits you can set a monthly spend limit in Settings > Usage. On Max and Ultra, the separate weekly Fable limit follows the same options: wait, enable usage credits, or switch models.
Why some people burn it in minutes
Anthropic's own Claude Code usage guide (fetched 2026-10-20) says almost every report of a limit gone before lunchtime traces back to the same habits: not clearing context between tasks and spending the most expensive model on work a small one would finish. Its countermeasures are matching the model to the job - Sonnet for routine changes, Opus for genuinely hard problems - and running /clear before a new task, because a long session resends everything it already carried on every turn.
The official error reference also states that one turn counts against the session and weekly allowances at the same time, so a single large fanout can empty the weekly allowance first. Public reports before that shared a few patterns: running several threads in parallel, letting the model sweep an entire repository, and spending the most expensive model on work a small one would finish.
Tightening scope, forbidding large refactors outright and pushing simple work to lightweight models are the fixes the community keeps confirming.
On QCode: billed per token, outside the subscription weekly limit
When you run out, first tell which side ran out: the messages above come from your Claude subscription, and QCode's API is billed per token and is not tied to the subscription weekly limit. On QCode, Personal Plans give a daily quota valid for 30 days, and if a day runs short you can buy a booster pack.
No Personal Plan on your account and not ready for a monthly one? Pay-as-you-go Packs start at $4 for a one-time quota you use within 7 days; when that quota runs out requests return HTTP 402, and after expiry HTTP 401. One key covers Claude, GPT, GLM, Kimi, DeepSeek and Qwen; per-model rates are listed on /models.
FAQ
How do I tell which cap I hit?
In Claude Code the message names it: "You've hit your session limit" is the 5-hour window, "You've hit your weekly limit" the weekly cap, "You've hit your Opus limit" or "You've hit your Sonnet limit" a single model family. Read the reset time in the message. A few hours out is usually the rolling 5-hour window; a few days out is the weekly cap. That is the fastest and most reliable signal. In Claude Code you can also run /usage, which lists your plan's limits and when each one resets. If the message starts with Fable limit reached, you hit the separate weekly limit that Max and Ultra give Fable. A verified Ultra account has no 5-hour window, so what it hits is the weekly limit or the Fable weekly limit; an Ultra account that is not verified yet has a temporary limit that is removed automatically once it is verified.
Can I reset the 5-hour window or the weekly limit myself?
Only with a limit reset, which Anthropic gives occasionally to eligible plans; otherwise both recover on their own and no support ticket is needed: the 5-hour window at the time in the message, the weekly limit once a week at a fixed time assigned to your account, both visible in /usage or Settings > Usage. Since v2.1.234 an interactive session signed in with a claude.ai subscription waits in place by default and resumes the task on its own at the reset. A limit reset puts the 5-hour or the weekly limit, whichever it covers, back to full right away; weekly limits still reset on their usual day, and Claude Code in the terminal or IDE has no entry for it - use the web or desktop app. To keep working before the reset, run /usage-credits or upgrade; a new plan's limits apply right away. Beyond that, treat any offer to instantly lift an official cap with suspicion.
Has the official 50% boost ended?
It ended. Help Center, fetched 2026-09-20 and rechecked 2026-10-08 and again 2026-10-20, is still past tense: the promo ran 2026-05-13 to 2026-09-13. From 2026-09-14 weekly limits are 25% above pre-promotion. Do not plan capacity on the +50%.
Why does one small fix eat a big slice of the weekly cap?
Usually because the model went out of scope: it swept the repo, questioned the architecture, refactored config, and only then touched the bug. Pinning the scope and forbidding large refactors normally helps a lot.
Does QCode have a 5-hour window too?
QCode does not count in hour windows. A Personal Plan is a daily quota that resets at 00:00 Beijing time and stays valid for 30 days; if a day runs short you can buy a booster pack, and booster packs reset separately at 02:00 Beijing time the next day, deliberately two hours later than the daily limit, leaving a buffer for work that runs past midnight. A Pay-as-you-go Pack is instead one total quota you use within 7 days; when it runs out requests return HTTP 402, and after expiry HTTP 401.
Will switching to QCode make this error go away?
For work you run through a QCode key, yes: it does not touch your official subscription allowance, because QCode's API is billed per token and is not tied to the subscription weekly limit; your Claude subscription keeps its own limits and reset times. The value is that when you run out you have other model families to switch to immediately.
Sources
Error text and limit rules: Anthropic Help Center https://support.claude.com/en/articles/11145838, https://support.claude.com/en/articles/12429409, https://support.claude.com/en/articles/15424964, https://support.claude.com/en/articles/17007452, https://support.claude.com/en/articles/15910845; Claude Code docs https://code.claude.com/docs/en/errors; release notes 2026-09-28 (Sonnet 5.5) and 2026-10-07 (Haiku 5.5); Help Center 11647753, 12429409, 17007452 and release notes 12138966. First fetched 2026-09-20, rechecked 2026-10-08 and 2026-10-20, rechecked again 2026-10-26; Help Center 15910845 records the +50% boost from May 13, 2026 to September 13, 2026, with weekly limits 25% above pre-promotion from September 14, 2026. Also checked on 2026-11-06: Help Center articles 11145838 and 17728952 (https://support.claude.com/en/articles/11145838, https://support.claude.com/en/articles/17728952) for Ultra, the wrap-up allowance and verified accounts, and the Claude Code error reference https://code.claude.com/docs/en/errors for the message text. Rechecked on 2026-11-09 for the limit messages, the wait banner, the Fable consent prompt and weekly resets: https://code.claude.com/docs/en/errors, https://code.claude.com/docs/en/interactive-mode, https://code.claude.com/docs/en/model-config and Help Center 11647753, 15424964 and 17007452. An October 2026 QCode check follows https://docs.qcode.cc/docs/reference/payg-packs and https://qcode.cc/pay-as-you-go.
Out of quota? Another model family is still there
One key across several model families; with a Personal Plan add a booster pack when quota runs short, without a plan Pay-as-you-go Packs start at $4 for 7 days.
Related reading
Prompt caching price tracker 2026
Cache read and write prices, TTLs and minimum cache lengths for Claude, GPT and DeepSeek, with two recomputable examples (as of 2026-10-07).
Claude Code cost optimisation
Concrete ways to spend the allowance where it counts.
Claude Code Pay-as-you-go Packs: No Subscription, 7 Days
Buy quota once and use it within 7 days, with no subscription; extra packs add to the same key.
Crypto payment guide
Nine coins, 56 networks, zero platform fee for AI coding API credits.
Official limits and the temporary boost follow Anthropic's published wording. The Help Center, fetched 2026-09-20 and rechecked 2026-10-08, puts the promo at 2026-05-13 to 2026-09-13 and weekly limits 25% above pre-promotion from 2026-09-14; that held on the 2026-10-20 recheck. Model availability follows /models. QCode's daily quota, booster packs and pack rules follow this site's pricing page and docs. Rechecked 2026-10-26: paid plans have a 5-hour session limit and a weekly limit, Max adds a separate Fable weekly limit, and usage credits are billed at standard API rates with a monthly spend limit. Rechecked 2026-11-06: Max and Ultra both have a separate Fable weekly limit, and verified Ultra accounts have no 5-hour limit.
Further reading
Pay-as-you-go Packs
One-time quota packs valid for 7 days, no subscription; extra packs add to the same key
Claude subscription refunds and account bans
As of 2026-10-13: the refund wording from Anthropic's terms and help center, the official request paths at Apple and Google Play with their timelines, and QCode
Claude invisible watermark
Text watermark + C2PA: a mark ≠ author, no mark ≠ not AI