Claude Code usage limits
What the 5-hour window and the weekly limit are on Anthropic's plans, when they reset, what Anthropic publishes and what it does not, and how many real coding tasks fit in a week on each plan, from our W38 2026 run.
On Claude Max 20x, Claude Fable 5.1 fits about 270 tasks a week through Claude Code. That is 14 passes of our fixed task set (real-world issues plus spec planning, 19 tasks) in one weekly window. A task is a whole multi-turn agent session, and the share it used is read off Anthropic's own meter.Measured
Provider pages reviewed Sep 14, 2026. Figures from the W38 2026 grading run, published Sep 14, 2026.
Two windows on one meter
Anthropic meters a Pro or Max subscription on two windows at once: a 5-hour session window and a weekly window. The pricing page sells the Max tiers as "5x or 20x more usage per 5-hour session than Pro" (claude.com/pricing, retrieved 2026-09-14). The Claude Code docs name the two ceilings a user hits as "You've hit your session limit" and "You've hit your weekly limit", and describe each as "a seat-based usage window on a subscription plan, shared across all models" (code.claude.com/docs/en/costs, retrieved 2026-09-14).
Our prober reads the same meter the CLI shows under /usage: a 5h window and a week window, each as a whole percent used, taken before and after every suite run. Every Claude Code row on the leaderboard is read on the weekly window, because that is the limit the plan is actually sold on. The meter also carries a third, per-model weekly cap ("You've hit your Opus limit" in the docs); we record it but do not grade on it.
When the limits reset
Both windows are rolling, not calendar-bound. The session window opens on your first request and closes five hours later. The weekly window opens with the first request after the previous one expired, and the meter reports its own reset instant for each window, which is what Claude Code shows when you hit a limit. Anthropic publishes no fixed reset weekday and no message or token count per plan.
A quiet account reads zero usage and no reset time until the first request opens the window. That is why our runs measure inside a declared quiet window: the delta from zero is the run's alone.
What a tier multiplies
The 5x and 20x multiples describe the session window. On the weekly window the gap is smaller: paired runs of the same models on Max 5x and Max 20x measured Max 20x holding about 2x the weekly allowance of Max 5x, while the session window grew about 4x as advertised. The table below scales by the weekly multiple where both tiers have one, and by the advertised session multiple otherwise (Claude Pro publishes no weekly figure we trust), exactly the way the leaderboard's budget picker does. Fable ships only on the Max tiers, so the Claude Pro row shows "not available" for it rather than a number.
Effort and model choice
The docs say that lowering the effort level with /effort reduces thinking tokens, and that "you can't turn off thinking on Fable models, which always use extended thinking" (code.claude.com/docs/en/costs, retrieved 2026-09-14). Anthropic publishes no figure for how much a higher effort setting consumes. Every Claude Code subject we grade runs at --effort medium, and we have not measured the same model at two effort levels, so this page carries no number for it either.
Model choice moves the meter too: the columns of the table below show how many passes of the same task set each model fits in one weekly window of the same plan.
Shared with Claude chat
Yes. Anthropic's support article states that "all activity in both tools counts against the same usage limits" (support.claude.com, retrieved 2026-09-14). A day of chat on claude.ai draws down the same weekly window your Claude Code tasks draw from.
How much a week holds on Claude Max 20x
Claude Max 20x's weekly limit holds about 7.1 fully spent 5h windows, out of the 33.6 a calendar week contains. Pooled over 908 runs in 31 session windows.
How many tasks fit in a week
One suite run is one pass of our fixed task set: real-world issues plus spec planning, 19 tasks. Each cell is how many of those passes one weekly window of the plan fits, from the share one pass consumed of the provider's own meter. A task is a whole multi-turn agent session on a real codebase, so this reads far lower than a message count would. Rows marked est scale the Claude Max 20x measurement by Anthropic's advertised usage multiples (claude.com/pricing, retrieved 2026-09-01), never by price; we have not run on those plans. Anthropic sells Max as 5x/20x the Pro allowance; our own paired 5x/20x runs measured ~5.7x where 4x is advertised, so the multiple can be well off.
| Plan | Claude Fable 5.1overall C | Claude Opus 5overall C | Claude Sonnet 5overall D | Claude Haiku 4.5overall F |
|---|---|---|---|---|
| Claude Pro$20/mo | not available | about 2.5 runsestabout 48 tasks | about 5 runsestabout 95 tasks | about 5 runsestabout 95 tasks |
| Claude Max 5x$100/mo | about 7.1 runsestabout 140 tasks | about 25 runsestabout 480 tasks | about 50 runsestabout 950 tasks | about 50 runsestabout 950 tasks |
| Claude Max 20x$200/mo · measured | about 14 runsabout 270 tasks | about 50 runsabout 950 tasks | about 100 runsabout 1900 tasks | about 100 runsabout 1900 tasks |
Hover or focus a cell for the derivation. "More than" marks a run the meter could not resolve: it read under one percent for the whole pass, so the count is a floor. A share is always of the plan's own allowance and is not comparable across providers.
The plans
Claude Pro at $20/month: not run on; its row is an estimate from the measured plan. Is Claude Pro worth it? · scorecards: Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5
Claude Max 5x at $100/month: not run on; its row is an estimate from the measured plan. Is Claude Max 5x worth it? · scorecards: Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5
Claude Max 20x at $200/month: the plan our subjects grade on, so its row is measured. Is Claude Max 20x worth it? · scorecards: Claude Fable 5.1, Claude Opus 5, Claude Sonnet 5, Claude Haiku 4.5
What Anthropic publishes
| Figure | Status |
|---|---|
| Session window length (5 hours) | published |
| A weekly window on top of the session window | published |
| Tier multiples per 5-hour session (Pro 1x, Max 5x, Max 20x) | published |
| Limits shared between Claude Code and Claude chat | published |
| Weekly multiple between tiers | not published |
| Message or token count per plan | not published |
| Reset weekday or hour | not published |
| Size of the per-model weekly cap | not published |
| Percent of each window one run consumed | our meter reads it |
Sources: claude.com/pricing, retrieved 2026-09-14; Using Claude Code with your Pro or Max plan (support.claude.com), retrieved 2026-09-14; Manage costs effectively (code.claude.com docs), retrieved 2026-09-14. Our own meter readings are described on each scorecard under "How the plan cost was measured".
Frequently asked questions
How many Claude Code tasks a week does Max 5x give you?
In one weekly window of Claude Max 5x, one pass of our fixed task set (real-world issues plus spec planning, 19 tasks) fits Claude Fable 5.1: about 7.1 passes (about 140 tasks), estimated; Claude Opus 5: about 25 passes (about 480 tasks), estimated; Claude Sonnet 5: about 50 passes (about 950 tasks), estimated; Claude Haiku 4.5: about 50 passes (about 950 tasks), estimated. Figures are estimated from our Claude Max 20x measurement by Anthropic's advertised tier multiples, W38 2026. A benchmark task is a whole multi-turn agent session, not a message.
How much of Claude Max 20x does one Claude Code task use?
About 0.4% of the weekly window on Claude Fable 5.1: one weekly window of Claude Max 20x fits about 270 tasks, read off the provider's own meter before and after each graded run, W38 2026. A benchmark task is a whole multi-turn agent session, not a message, so a provider's message estimate is a different unit.
When does the Claude Code weekly limit reset?
Seven days after the first request that opened it, not on a fixed weekday. The 5-hour session window works the same way: it opens on your first request and closes five hours later. Claude Code shows the exact reset time when you hit either limit, and Anthropic publishes no schedule beyond that (code.claude.com/docs/en/costs, retrieved 2026-09-14).
Does effort high burn the Claude Code limit faster?
Anthropic says a lower effort level reduces thinking tokens, which count against the limit, but publishes no multiple for effort high versus medium. We grade every Claude Code model at effort medium and have not measured the same model at two effort levels, so we state no number. The table on this page shows how far the model choice alone moves the weekly meter on one plan.
Is the Claude Code weekly limit shared with Claude chat?
Yes. Anthropic's support article on using Claude Code with a Pro or Max plan says all activity in both tools counts against the same usage limits (support.claude.com, retrieved 2026-09-14). Chat on claude.ai and tasks in the CLI draw from one 5-hour window and one weekly window.
Other harnesses: Codex CLI · Antigravity CLI. The leaderboard shows the same usage per plan-price band; compare subscriptions puts the plans side by side.