Claude Code Usage Limit Reached? Here's What to Do

Nothing is more disruptive than a mid-task lockout. Claude Code stops responding, and the error message tells you almost nothing useful: it says you've hit a limit, not which one, not how close you were, and not when it resets. This guide fills in every gap.

What exactly does “usage limit reached” mean?

Claude Code enforces usage through two layered windows, not a single hard cap. The first is a rolling 5-hour window that refreshes automatically once 5 hours have elapsed from your first request in the window. The second is a weekly aggregatethat resets once per calendar week. Hitting either one produces the same lockout message, which is one reason the error is so confusing — you don't know if you need to wait five hours or five days.

Why the error message is deliberately vague

Anthropic has stated that exact token counts per window are not published in the product, because they vary by plan tier (Pro vs Max), can change over time, and are intentionally left as soft targets rather than sharp cliffs. In practice this means the error gives you no number — just a wall. The only way to understand your utilization is to read it from your local logs.

What a locked-out session looks like

$ claude "refactor the auth module"   Claude Code  ✗ Usage limit reached    Your account has hit its usage limit.    Please wait before sending more messages.   (request blocked — no tokens consumed)

How do I find out what I actually burned?

Claude Code's built-in /usage command shows your current-window utilization as a percentage, but it does not break down which conversations consumed the most, and it resets when the window rolls over. To get a durable picture — across sessions, days, and tools — read the underlying logs directly.

$ npx whoburnedmore  ↳ scanning claude code logs…   DATE         TOKENS IN    TOKENS OUT   COST  2026-06-13   6.2M         1.1M         $24.30  2026-06-12   4.9M         0.9M         $19.40  2026-06-11   3.1M         0.6M         $12.20  ─────────────────────────────────────────  weekly       14.2M        2.6M         $55.90

The day you hit the limit almost always jumps out immediately. In the example above, June 13 burned nearly double a normal day — that single session filled the window. Knowing this lets you plan: next week, keep days under 5M input tokens.

When does the limit reset?

The reset timing depends on which window you hit. The table below is the fastest way to reason about your situation:

1. 1

   Check whether /usage shows >80% inside a new session

   If Claude Code starts a session at all, run /usage immediately. A near-full bar means you burned your 5-hour window and it has not cleared yet — try again once the 5-hour mark passes from your last heavy request.
2. 2

   If the session never starts, suspect the weekly cap

   Weekly caps reset at a fixed calendar point (usually midnight UTC on Monday, though Anthropic has not published an exact schedule). If five hours have clearly passed and you're still blocked, wait for the weekly reset.
3. 3

   Check your timestamps in whoburnedmore output

   The per-day breakdown shows exactly when your biggest burn happened. Add 5 hours to that timestamp; if it's still in the future, that's your 5-hour gate. If it's in the past and you're still blocked, it's the weekly cap.
4. 4

   Use the quiet window productively

   Write tests, review diffs, and draft prompts offline. You can also switch to npx whoburnedmore --local to open a local usage dashboard and study your consumption patterns without touching any Claude API.

How do I avoid hitting the limit again?

The limit is a symptom of consumption you didn't see coming. The fix is better visibility before you hit the wall, not damage control after. A few habits matter more than any individual optimization.

Keep an eye on your daily burn rate

Run npx whoburnedmore at the end of each working day. Once you know that a normal day costs you 3–4M input tokens, you know that a spike to 8M is two days of budget in one. Catching that pattern on Tuesday prevents the Friday lockout.

Shorten your context before expensive tasks

Every token in your context window is billed against your limit, even if it's a file Claude has already read three times. Run /compact before starting a long refactor, or /clear to reset the context entirely. A 200K-token context costs 200K tokens on every message, so trimming it to 50K before a 10-message session saves 1.5M tokens.

Cross-link your usage data across tools

If you also run Codex CLI, Gemini CLI, or Cursor, their token burns are completely separate from Claude Code's quota — but they come out of the same developer hours. Seeing the full picture in one place (via npx whoburnedmore) helps you decide which tool is giving you the best tokens-per-insight ratio. For a deep dive on how Claude Code's limits compare to Codex's, see the Claude Code rate limits explainer.