Claude Code Usage Limits, Explained: The 5-Hour Window and the Weekly Cap

Published September 23, 2026 · by the AQ team

Claude Code runs on your Claude subscription, and as of September 2026 that subscription carries two stacked usage limits: a session limit that resets every five hours, and a weekly limit that resets at a fixed time Anthropic assigns to your account. Both are shared across every Claude surface, so tokens spent in Claude Code, the claude.ai chat, and the desktop app all draw from one pool. Anthropic no longer publishes exact token allowances; plans are sized relative to Pro (Max is "5x" or "20x" more per session), and the fastest way to see your real remaining budget is to type /usage inside a Claude Code session. This guide explains how each limit works, what drains them, how the numbers changed during 2026, and your options at the limit.

How does the 5-hour session limit work?

A session is a rolling five-hour usage window: it opens when you send your first message, everything you do for the next five hours draws from that window's allowance, and per Anthropic's help center the session limit "will reset every five hours." Two consequences follow. The reset time depends on when you started, not on the clock: begin at 8:07am and the window resets around 1:07pm. And a session is a usage budget, not a time budget: light use for five hours may barely dent it, while one heavy agentic run can exhaust it in under an hour.

On May 6, 2026 Anthropic permanently doubled Claude Code's five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans, and removed the peak-hours limit reduction that previously throttled Pro and Max accounts during busy periods.

How does the weekly limit work?

The weekly limit is the ceiling above the sessions: a budget across all models that resets once a week at a fixed day and time assigned to your account (visible in /usage, or Settings then Usage on claude.ai). Anthropic introduced weekly limits on August 28, 2025, estimating they would affect fewer than 5 percent of subscribers, mostly people running Claude Code around the clock.

The weekly numbers moved twice in 2026. A temporary boost from May 13, 2026 raised weekly limits roughly 50 percent above the standard level. That promotion ended September 13, 2026, replaced on September 14 by a permanent increase of about 25 percent over the old standard. Net effect: compared with August 2026 you have about 17 percent less weekly headroom, a reduction Anthropic acknowledged in its follow-up. If your weekly budget suddenly felt tighter in mid-September 2026, that change is why, not a bug in your setup.

What actually draws down the limit?

Usage is metered in tokens processed, and per Anthropic's help center it scales with message length, attachment size, current conversation length, tool usage, model choice, and effort level. In Claude Code the expensive part is rarely your prompt: every turn resends the conversation so far, your CLAUDE.md project context, and every file Claude has read this session, then adds the model's output, tool results, and thinking on top. That is why a long-lived session drains the meter faster and faster: the context it re-processes each turn keeps growing. Auto-compaction softens this, but a fresh session per task remains the single cheapest habit.

Model choice is the other big lever. Opus-class models consume the shared budget several times faster than Sonnet-class models for the same work, which is why heavy users default to Sonnet and reserve Opus for problems that need it. Anthropic's original weekly-limit sizing from August 2025 gives a feel for the ratio: Pro was estimated at 40 to 80 hours of Sonnet per week, while the $100 Max plan got 140 to 280 Sonnet hours but only 15 to 35 Opus hours. Those figures predate the 2026 changes, but the rough 5-to-1 spread between tiers still holds.

What each plan includes, as of September 2026

All figures from Anthropic's pricing page, September 2026. The multipliers describe per-session capacity relative to Pro; Anthropic does not publish absolute token counts per plan.

How do I check how much usage I have left?

Inside a Claude Code session, type /usage. It draws progress bars for your current five-hour session and your weekly limit, with the reset time for each, and reading it does not consume any usage. On claude.ai, Settings then Usage shows the same two meters. The separate /cost command tracks dollar spend for people running Claude Code on API-key billing rather than a subscription. Claude Code also warns you in the session as you approach a limit.

What happens when you hit a limit?

Claude Code stops accepting new prompts and tells you when the limit resets. Nothing is lost: the conversation history and your working tree stay intact, and the session resumes when the window reopens. From there you have four options. Wait for the reset (at most five hours for a session limit). Switch to a cheaper model with /model, since Sonnet stretches the same budget several times further than Opus. Enable usage credits: on Pro and Max plans, Settings then Usage lets you prepay a balance that kicks in after your included usage runs out, billed at standard API rates with a $2,000 daily redemption cap, without changing your reset schedule (credits do not expire, except in certain jurisdictions such as Japan, where credits bought on or after September 10, 2026 expire after six months). Or upgrade the plan outright.

Scheduling work around the windows

Teams that live inside these limits converge on the same habits:

• Start long runs at the top of a window. A five-hour window you open at 9am is gone by 2pm regardless of what you did with it. Kicking off the big refactor right after a reset, instead of 20 minutes before one, is free capacity.
• One task per session, one session per worktree. Context accumulation is the silent drain. A fresh session in an isolated git worktree starts from zero context instead of paying for the last task's leftovers on every turn.
• Parallelize tasks, not marathons. Three focused sessions on three worktrees finish more work per token than one nine-hour session, because none of them re-reads a bloated history. Running multiple Claude Code sessions in parallel covers the mechanics; parallel sessions do all draw from the same account pool at once.
• Keep the machine running. Limits are account-scoped, not machine-scoped, so a server buys no extra tokens. What it buys is that the usage you have lands in finished work: a run that survives a closed laptop uses the window you paid for instead of dying with your SSH connection.
• Split people, not accounts. Each teammate's subscription is its own pool, so a team's real capacity is per person. Some individuals go further and juggle plans; running multiple Claude Code accounts explains when that is worth it.

When does API billing beat a subscription?

As of September 2026, Anthropic's API lists Claude Sonnet 5 at $2 per million input tokens and $10 per million output tokens (the scheduled increase to $3/$15 was canceled), and Claude Opus 5 at $5 and $25. A subscription is a bulk discount with a ceiling; the API is metered with no ceiling. Steady daily interactive use almost always favors the subscription; the API wins for spiky automation such as CI jobs, batch refactors, or an occasional very heavy week where the cap would stall you. The pragmatic middle path is usage credits, which bill overflow at exactly API rates: one month with them enabled tells you precisely what uncapped API billing would cost. Coding agent token costs for teams goes deeper on budgeting this across a team.

Where AQ fits

AQ is the multiplayer coding harness where engineering teams run AI coding agents like Claude Code and Codex together: shared live terminals, a code editor, and app previews, in your own cloud. AQ does not change your Claude limits, and does not resell model access: each person connects their own Claude account through a per-user CLI login, and AQ never marks up usage on your own subscription. What it changes is how much of your allowance becomes finished work. Sessions run in persistent tmux on your team's VM, so the run you start at the top of a five-hour window keeps working after your laptop closes and is still there, resumable from any device, when the window resets. Each workspace gets its own isolated git worktree, which makes fresh-session-per-task the default rather than a discipline, and teammates can open the same live session, so a run that hits the weekly cap on one person's account can be picked up by another teammate on their own. The Free plan is a personal sandbox AQ creates for you with nothing to install and no time limit; the Team plan is $50 per user per month in early access (standard $200), covering VMs you connect from your own cloud or a dedicated always-on AQ-managed VM, with the rate locked for your first 12 months.