When does the Claude Code weekly limit reset?

The weekly limit is a rolling seven-day window, not a calendar week. Claude Code shows your exact reset time in the /usage command and in the message you see when you hit the limit. The 5-hour session limit is separate and resets five hours after the first message of the session.

Why did I hit my weekly limit so fast?

Almost always long context. Claude Code sends your entire conversation with every request, and every tool call is another request. A one-line question in a session that's been open all day carries the whole day's history. Cache misses after a break longer than the cache lifetime (an hour on a subscription) reprocess your full context, and Opus draws the allowance about 2.5 times faster than Sonnet. Clearing between tasks fixes most of it.

Is the Opus limit separate from the weekly limit?

On Max plans, yes. There's an all-models weekly cap and a separate, smaller weekly Opus allowance (roughly 15–35 hours on Max 5x, 24–40 on Max 20x). If you see 'You've hit your Opus limit', switching to Sonnet with /model keeps you working; if you see 'You've hit your weekly limit', it applies to every model.

What can I do when I hit the limit?

Four options: wait for the reset (Claude Code can wait and continue the interrupted task automatically), switch models if it was a model-specific limit, turn on usage credits to keep working at API rates up to a spend limit you set, or upgrade the plan. Pro is 40–80 Sonnet hours a week; Max 5x is 140–280.

Claude Code Weekly Limit Explained: How It Works, When It Resets, and How to Stretch It (2026)

Three limits, not one

Limit	Applies to	Resets	Message you'll see
5-hour session window	All models, all Claude surfaces on the plan	5 hours after the first message of the window	"You've hit your session limit"
Weekly cap	All models; shared with Claude chat and Cowork	Rolling 7 days; exact time shown in `/usage`	"You've hit your weekly limit"
Weekly Opus allowance (Max only)	Opus-family models only	Rolling 7 days	"You've hit your Opus limit"

The three are independent. Running out of the 5-hour window doesn't touch the weekly cap, and vice versa. The message matters: a model-specific limit ("Opus limit", "Sonnet limit") means switching to a model outside that family with /model keeps you working. A session or weekly limit is shared across all models, so switching does nothing.

Published sizes, per Anthropic's plan page: Pro gets roughly 40–80 Sonnet hours a week; Max 5x 140–280 Sonnet plus 15–35 Opus; Max 20x 240–480 Sonnet plus 24–40 Opus. The weekly limit calculator estimates which day of the week you'll run out from your hours and session intensity.

Why it's measured in hours but spent in tokens

Anthropic describes the limits in "hours of Sonnet" because that's intuitive, but the meter underneath counts tokens. That's why the published figures are ranges: an hour of short prompts on small files might cost a fifth of an hour spent in a 150,000-token conversation with tool calls flying. Four things move you between the low and high end of the range:

Context length. Every request resends the whole conversation. Prompt caching makes re-reading history cheap (cached tokens cost a tenth of fresh ones) but not free, and a session open since morning carries the morning with it.
Cache misses. After a break longer than the cache lifetime, an hour on a subscription, your next message reprocesses the full context at full price. The /usage breakdown flags "cache misses" when they're 10% or more of your recent usage.
Model. Opus costs about 2.5 times Sonnet per token and Fable about 5 times, so an Opus session draws the shared weekly cap proportionally faster, as well as the Opus allowance.
Parallelism. Subagents, agent teams and scheduled tasks each send their own context. Anthropic's docs put agent teams at roughly seven times the tokens of a normal session.

Things that use your limit while you're not typing

This is the part that surprises people. A session can draw usage while idle: scheduled tasks fire on their interval and send the full context each time; cross-session messages from another Claude Code session arrive as new turns; goal check-ins on background work start new turns (capped at three per goal); active agent teammates keep consuming until they exit. Background summarization for claude --resume uses a small amount too, usually under $0.04 of API-equivalent per session. If your weekly bar moved overnight, look for a loop or a teammate you forgot about. /usage lists the heaviest scheduled tasks by tokens.

How to see exactly where you stand

/usage shows plan usage bars for the 5-hour and weekly windows, a breakdown of what's consuming them (skills, subagents, MCP servers, loops), behavior flags like long context, and your usage-credit spend if credits are on. Press d or w to toggle 24 hours vs 7 days.
/context shows what's occupying the current context window, which is where the per-request cost comes from.
The status line can display context usage continuously so you notice a bloated session before it eats a window.
On claude.ai, Settings → Usage shows the same bars plus usage-credit balance and spend limit.

What to do when you hit it

Read the message. Model-specific: switch models. Session: wait up to 5 hours. Weekly: wait for the shown reset time.
Let Claude Code wait for you. Recent versions can hold the interrupted task and continue automatically when the window resets (/rate-limit-options).
Turn on usage credits. /usage-credits opens Settings → Usage, where extra usage bills at standard API rates up to a monthly limit you set. Note the cache lifetime drops from an hour to five minutes while you're on credits, so long idle gaps get more expensive.
Upgrade if it's the third time this month. The Pro vs Max guide has the hours math.

How to stretch it

In order of impact: /clear between unrelated tasks (with /rename first so you can /resume later); keep Sonnet as the default and reach for Opus deliberately; keep CLAUDE.md under 200 lines and move workflow instructions into skills that load on demand; disable MCP servers you aren't using; delegate log-reading and test runs to subagents so the verbose output stays out of your main context; and write specific prompts, because "improve this codebase" triggers a scan of everything. The token-saving guide goes through each with the command to run.

Estimate the day I'll hit my weekly limit →

Frequently asked questions

When does the weekly limit reset?

On a rolling seven-day cycle. /usage and the limit message show the exact time.

Is the 5-hour limit part of the weekly limit?

No. They're independent windows; you can exhaust one without touching the other.

Is the Opus limit separate?

On Max plans, yes: a smaller weekly Opus allowance alongside the all-models cap. Switching to Sonnet gets around an Opus-only limit.

Why did an idle session use my limit?

Scheduled tasks, agent teammates, cross-session messages and goal check-ins all send your full context without you typing.

Does Claude chat count against the same limit?

Yes. Chat, Cowork and Claude Code share the plan's allowance.

Can I buy more without upgrading?

Yes: usage credits bill extra usage at API rates up to a monthly limit you set.

The Claude Code weekly limit, explained