Claude Code usage limits, explained — the 5-hour window, weekly caps, and how to see your burn

Quick answers

Is there a daily limit? No. There’s a 5-hour session limit that “will reset every five hours,” plus weekly caps on top that “reset seven days after your session starts” — rolling, not calendar-day.

Do Claude chat and Claude Code share one pool? Yes, verbatim from the docs: usage is “shared across Claude and Claude Code, meaning all activity in both tools counts against the same usage limits” — claude.ai, Desktop, and the CLI all draw from the same meter.

What happens when I hit it? A warning (“Approaching 5-hour limit”), then a blocking message with your reset time: “5-hour limit reached — resets [time].” No model downgrade; you wait, or overflow with Usage credits.

Can I pay to keep going? Yes — Usage credits (the feature formerly called “extra usage”): past your subscription limit, work continues “billed at standard API pricing rates,” with a monthly spend cap you control.

How the meter actually works

Two clocks run at once:

1. The 5-hour session window. Anchored to when your session starts; it resets every five hours. The docs don’t define the trigger more precisely than that — treat “first message starts the clock” as folklore, close but not official.
2. Weekly caps on top. Pro has one (across all models); Max and Team Premium seats have two — “one that applies across all models and another for Sonnet models only”. Both reset seven days after your session starts.

The practical consequence of the shared pool: a heavy Claude Code afternoon also empties your chat allowance, and vice versa. Multi-surface users hit walls that single-surface users never see — that’s the meter working as documented, not a bug.

What changed on May 6

Anthropic doubled Claude Code’s five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans, and removed “the peak hours limit reduction on Claude Code” for Pro and Max. If your sense of the limits formed before May, it’s stale in your favor — the same money buys roughly twice the agent time, and evening sessions no longer get quietly squeezed.

What each plan actually gets

Official multipliers, all relative to Pro:

That 6.25× is the least-known number in the lineup: a Team Premium seat sits a notch above Max 5× on session usage at the same sticker price, with admin-managed overflow instead of self-serve. Which tier pays off for you is a pricing question, not a limits question — this table is just the quota side of it.

See your burn before the wall

• /usage in Claude Code — plan usage bars plus a breakdown that attributes recent usage to skills, subagents, plugins, and individual MCP servers. Press d/w to flip between 24-hour and 7-day views. One documented caveat: figures are “computed from local session history on this machine,” so other devices and claude.ai usage don’t show here.
• /status — quick check of remaining allocation; warnings appear here as you approach the wall.
• Web: Settings → Usage — progress bars for the 5-hour session and weekly limits, the authoritative cross-device view.
• Status line — configurable to show context-window usage continuously; note that’s context fill, not your plan percentage.
• Community tooling exists because this anxiety is real: ccusage analyzes token usage and cost from your local session files, and Claumon turns them into a live dashboard. Useful, with the same local-only blind spot as /usage. (That this is the densest cluster in Claude Code’s accessory ecosystem is itself a verdict on native limit visibility.)

The workflow that prevents most lockouts: glance at /usage before starting anything fleet-shaped, and check the weekly bar — the 5-hour wall is annoying, but the weekly wall is the one that ends your day.

When you hit it anyway

The sequence is documented: warning first, then a blocking “5-hour limit reached — resets [time].” Your options:

• Wait. The message includes your reset time; weekly resets are shown in Settings → Usage.
• Overflow with Usage credits. Enable in Settings → Usage (Pro and Max self-serve; Team/Enterprise request it from your admin via /usage-credits). Past the limit, the lockout message becomes “continuing with usage credits” and you’re billed at standard API rates — with a monthly spend limit you set, optional auto-top-up, and a $2,000 daily redemption ceiling. In the CLI, /usage-credits sets the cap and prompts you to raise it mid-session instead of stopping. (If you knew this feature as “extra usage” — same thing, renamed; the old /extra-usage command still works.)
• Re-plan the work. If overflow bills become routine rather than crunch-day exceptions, the tier math has flipped and the upgrade is cheaper — and if you’re regularly burning sessions on babysitting rather than output, the fix is usually workflow design, not quota.
• Shrink the output itself. Over-built code burns quota twice — once to generate it, once to review and revise it. Code-volume rules in CLAUDE.md are the cheapest lever here; the benchmark behind the viral version self-reports 47% fewer tokens on identical tasks.

Companion reading

• Claude Code pricing — which tier the math actually favors.
• Best AI coding agents, June verdict — where these limits sit in the cross-vendor picture.
• Running a fleet of long-lived agents — the usage pattern most likely to meet these walls.
• Managing Claude Code’s context window — cut the per-turn token burn that pushes you into these caps faster.
• Codex vs Claude Code usage limits — the same limit machinery, and why the wall arrives sooner on one side.