Claude Code's Weekly Limit, Explained (and How to See It Coming)

The 5-hour limit is annoying. The weekly limit is the one that actually ruins your week: you wait out the usual break, come back, and you are still locked out, sometimes for days. This post explains how the weekly limit actually works, how to tell which limit you hit, why your usage drains faster than your activity suggests, and every way to see it coming before it lands.

Disclosure: I work on Blume, a desktop app mentioned in the tools section. Everything else here comes from Anthropic's documentation and the built-in commands.

The two limits, in one minute

Claude Code subscriptions (Pro, Max, Team, Enterprise) enforce two overlapping windows:

1. A rolling 5-hour window. Limits the pace of your usage. Wait, and it comes back.
2. A weekly window. Caps your total volume. When this one is gone, waiting an hour does nothing.

Two details cause most of the confusion:

• The pool is shared across products. Claude Code, claude.ai chat, and Cowork all draw from the same allowance. A long brainstorming session in the web chat eats the same budget your coding sessions use.
• It is shared across models too. Switching from Opus to Sonnet with /model keeps you working after an Opus-specific limit message, but it does not restore access once you have hit the overall session or weekly limit.

Which limit did I hit?

The error message tells you: "You've hit your session limit" is the 5-hour window, "You've hit your weekly limit" is the weekly cap. Both show when the window resets. The practical test, if you missed the message: if a short break restores access, it was the 5-hour window. If you stay blocked no matter how long you wait, it was the weekly cap.

Run /usage at any time to see both bars, your reset times, and activity stats. On claude.ai, Settings > Usage shows the same thing in a browser.

Why the weekly limit drains faster than you expect

This is the part most guides skip, and it is where the budget actually goes. Claude Code sends your full conversation with every request, so context size, not message count, drives consumption. Anthropic's own docs list the main culprits:

• Long-lived sessions. A one-line question in a session that has been open all day still carries the whole conversation. Prompt caching makes re-reads cheaper, but they are not free.
• Cache misses after breaks. On a subscription, the prompt cache lives for about an hour. Come back from lunch and your first message reprocesses the full context at normal rates. (Once you are drawing on usage credits, the cache lifetime drops to five minutes.)
• Compaction. /compact reads everything it summarizes, so compacting a huge context is itself a huge request. /clear costs nothing, so prefer it when you do not need continuity.
• Scheduled tasks and agent teammates. Anything that fires on an interval, or any parallel teammate you forgot to shut down, keeps drawing from the pool while you are not looking. Agent teams can use roughly 7x the tokens of a normal session.

The newer builds of Claude Code surface this directly: the /usage breakdown now attributes recent usage to skills, subagents, plugins, and individual MCP servers, and flags behaviors like long context or cache misses when one accounts for 10 percent or more of your recent usage. Press d or w to toggle between the last 24 hours and the last 7 days. If you have not looked at this screen since it shipped, it answers "where did my week go" better than any external tool used to.

Seeing it coming: the tooling options

Built-in: /usage for the bars and the attribution breakdown, /context for how full the current session is, and a status line if you want context usage always visible.

CLI reports: ccusage reads your local session logs and produces daily, weekly, and 5-hour-block reports, useful for understanding patterns across weeks. Claude Code Usage Monitor gives you a live terminal view with burn-rate predictions for the current window.

Desktop: Blume shows plan usage for Claude Code and Codex side by side, next to a live view of what every agent is doing (working, finished, or waiting on approval). The angle is different from the CLI tools: usage spikes are often a symptom of setup problems, like an agent stuck re-reading files because its instructions are unclear, and Blume also tracks the rules, skills, and hooks steering each agent, with automatic mismatch detection on its roadmap. If you only run Claude Code in one terminal, the built-ins plus ccusage cover you; Blume earns its place when several agents share your weekly budget.

What to do when you hit the wall

• Usage credits. Paid plans can now continue past the limit with usage credits: run /usage-credits to set it up. Older "you just have to wait" advice predates this.
• Make the budget go further. Clear between unrelated tasks, resume big sessions from a summary instead of the full history, use Sonnet for routine work and save Opus for the hard problems, lower the thinking effort for simple tasks, and move rarely-used instructions out of CLAUDE.md into skills that load on demand.
• Check the breakdown before changing habits. If /usage says one MCP server or a cache-miss pattern accounts for a third of your consumption, fix that first. One config change can be worth more than a week of careful prompting.

FAQ

Does claude.ai chat count against my Claude Code weekly limit?
Yes. Pro and Max usage is one pool shared across Claude Code, claude.ai, and Cowork.

When exactly does my weekly limit reset?
The limit message and /usage both show your reset time. Do not rely on blog posts for this one: third-party write-ups disagree about the reset mechanics, and your own /usage screen is authoritative for your account.

Does switching models restore access?
No. Session and weekly limits are shared across all models. Switching only helps with model-specific messages like the Opus limit.

Can I pay to keep working?
Yes, with usage credits on paid plans, or by using an API key with pay-as-you-go billing, which has no weekly cap but bills per token.

Do background processes really use my quota?
A little. Conversation summarization and some status commands consume small amounts (typically under $0.04 per session), but scheduled tasks and idle agent teammates can consume a lot more.

If your weekly limit keeps surprising you, the /usage attribution screen is the first thing to check. And if you have found a consumption pattern that surprised you, share it in the comments: the failure modes are more varied than the docs suggest.