Claude Code Usage Limits, Explained

Limit one: the rolling 5-hour session

Every message you send starts or joins a 5-hour window. Your plan includes a certain amount of usage per window, shared across Claude Code, claude.ai, and the desktop and mobile apps. When the window expires, a fresh one starts with your next message — so "wait for the reset" is never more than a few hours away. In May 2026 Anthropic permanently doubled the 5-hour limits on Pro, Max, Team, and seat-based Enterprise plans, which moved the everyday pain from session limits to the weekly cap.

How much a session "costs" depends heavily on what you do with it: model choice, codebase size, and how much context each request drags along. The same hour of work can consume wildly different amounts of your allowance — which is why measuring your own usage (below) beats rules of thumb.

Limit two: the weekly cap

On top of sessions sits a weekly limit covering all your usage across all models. Unlike the session window, the weekly reset is fixed, not rolling — it resets at a set time each week assigned to your account, whether you've been coding or not. This is the limit heavy users actually plan around: burn hard early in your week and no amount of session discipline gets it back before the reset.

Checking where you stand

Three tools, three different questions:

• /usage in Claude Code (or claude.ai/settings/usage) — am I about to hit a limit, and when does it reset? This is the official percent-used view of both the session and the week.
• npx ccusage@latest daily — what am I actually consuming? It reads your local session logs and computes per-day tokens and API-equivalent USD cost, per model. This is the number that tells you whether your subscription is a bargain. Full guide here.
• npx viberank-cli — is my usage normal? It submits your ccusage totals to the public leaderboard, where you can compare against 1,100+ developers. Median heavy users burn far more than most people guess.

One number from the board worth knowing: across 11+ trillion tracked tokens, about 95% are prompt-cache reads that cost roughly a tenth of normal input tokens. A scary token count in ccusage usually maps to a much smaller real cost — and to less limit pressure than you'd fear.

Hit the limit? Your options, in order

1. Wait it out. Session windows refresh within hours; /usage shows your weekly reset time.
2. Burn less per task. Route routine work to smaller models, keep context lean, start fresh conversations instead of dragging 100k-token histories. Nine concrete tactics here — the same habits that cut API bills stretch subscription limits.
3. Enable usage credits. Pro and Max plans can keep working past the included limit at standard API rates, billed pay-as-you-go — no plan change required.
4. Switch to an API key for the overflow. API billing has no session or weekly caps, just per-token pricing and org rate limits.
5. Upgrade — but check the math first. Max 5x ($100/mo) and Max 20x ($200/mo) multiply your limits, and for heavy users they're dramatically cheaper than equivalent API usage. Whether you're a heavy user is an empirical question: drop your ccusage numbers into the calculator and see which tier your actual usage justifies, or read our full Pro vs Max breakdown.

The subscription arbitrage, quantified

Limits exist because flat-rate plans are a genuine arbitrage. On the Claude Code leaderboard, developers routinely show thousands of dollars of API-equivalent usage per month on a $100–$200 subscription — the top of the board is past six figures in tracked equivalent spend. If you keep slamming into weekly caps, you're almost certainly in the population for whom Max pays for itself many times over.

One command puts your real numbers on the board — your code and prompts never leave your machine, only usage totals do.

Claude Code limits FAQ