If you use Claude Code on a Pro or Max subscription, you have probably seen the message that you have reached your limit halfway through a task. The limits are not hard to understand, but they are easy to lose track of, because they are shared, they reset on different schedules, and Claude Code burns through them faster than chat.
The short version
When you sign in to Claude Code with a subscription, it draws from the same limits as claude.ai:
- a rolling 5-hour session that starts with your first message,
- a weekly limit shared by every model,
- and, on some plans, a separate weekly cap for specific models.
Chat and coding share one pool. A long refactor in the terminal can leave you blocked on claude.ai, and a long chat can cut a coding session short.
If you use Claude Code with an API key from the Claude Console instead, none of this applies: you pay per token under your API rate limits.
| How you use Claude Code | Limits | Where to see them |
|---|---|---|
| Pro or Max subscription | Shared 5-hour session and weekly limit with claude.ai, plus per-model weekly caps |
/usage in Claude Code, claude.ai/settings/usage |
| API key | No session or weekly limit; pay per token within your rate limits | The Claude Console |
Why Claude Code uses the limit faster than chat
- Agentic loops. One request can turn into many model calls while Claude Code reads files, runs commands and edits code.
- Large context. Each call carries the conversation so far plus every file it has read, so long sessions cost more per step.
- Stronger models. The most capable model uses more of your limit per message than a lighter one.
How to make it last
-
/clearbetween unrelated tasks and/compactto shrink a long conversation, so each call carries less context. -
Match the model to the job with
/model, and keep the strongest one for the hard parts. - Plan around the reset. If the session window is nearly used up and resets in 40 minutes, start the big refactor after the reset, not before.
-
Keep the numbers in view.
/usageshows your session and weekly usage with reset times, but it has no history and does not warn you before you run out, so it is easy to forget mid-task.
Tools that help
-
/usage(built in): quick check of the session and weekly percentages. - ccusage (open source): reads Claude Code's local logs and reports tokens and estimated cost per day, session or 5-hour block. It answers "how much did I use".
- Claude Monitor (open source, I'm the author): a browser extension that keeps the official session and weekly percentages, per-model caps, extra usage credits and Claude Code routine runs in your toolbar, with a countdown to each reset and alerts at 80% and 95%. It answers "how much is left", including what you used in claude.ai.
The longer version, with a FAQ, is on claude-monitor.com. If you have a trick for stretching the limits that I missed, I would like to hear it in the comments.
Top comments (0)