Claude Code’s ‘25% Bigger’ Weekly Limits Are a 17% Cut

Open Claude Code today, glance at the usage bar, and everything looks healthy: you are running on a weekly limit that is 50% higher than the standard one, courtesy of a temporary boost Anthropic has had in place for a while. On 29 August, Anthropic told users that arrangement is changing. Its message: from 14 September, it is permanently raising the standard weekly limits by 25% for Pro, Max, Team and seat-based Enterprise plans. Read quickly, that sounds like good news — more headroom, made permanent.

Read slowly, and it is a cut. The temporary 50% boost expires on the same day the permanent 25% increase begins. Put a number on it: if the pre-promotion baseline was 100, today you are on 150, and on 14 September you drop to 125. Going from 150 to 125 is a reduction of about 17%. So the same subscription buys roughly a sixth less weekly Claude Code usage next month than it does this month — while, against the older baseline, buying 25% more than it once did. Both figures are real. The question is which one you lead with.

Anthropic led with the increase. The reduction is the number that describes your actual, lived allowance from one month to the next, and it arrived second — and, according to reporting, only after the first message was taken down and rewritten. That is the story: not that Anthropic did something unusually harsh, but that a company chose to announce a cut by putting the word “raising” at the front of the sentence.

What Anthropic actually announced

The wording is worth quoting, because both numbers come from Anthropic itself. The first post said: “Starting September 14, we’re permanently raising standard weekly limits in Claude Code by 25% for Pro, Max, Team, and seat-based Enterprise plans,” adding that “until then, the current 50% increase will be in place.” Everything you need to work out the cut is right there — a 25% permanent rise, a 50% temporary boost ending simultaneously — but the arithmetic that turns those into “about 17% less than today” is left for the reader to do.

As BleepingComputer reported, Anthropic then deleted the original thread and posted a clarification that does the arithmetic out loud: “Compared to today, this works out to a 17% reduction in weekly limits on Claude Code.” The reposted message continued: “We’re working on exciting changes that will make it feel like you’re getting more from Claude, while having more visibility and control of your usage.” Credit where it is due: the clarification is unambiguous, and Anthropic spelled the cut out in its own words rather than leaving it to be discovered. But the sequence — increase first, reduction second, after a delete-and-repost — is the part users noticed.

The case for Anthropic, made honestly

There is a real defence here, and it deserves stating before the criticism. The permanent limit genuinely is higher than the pre-promotion baseline; a user who joined before the boost and stays past September ends up with more than they started with. The 50% boost was always labelled temporary, and a promotion ending is not the same as a product being downgraded — a point some users made in Anthropic’s defence. And running frontier models at a flat monthly price is expensive in a way that genuinely constrains how much usage a subscription can include; Anthropic is not obliged to keep a promotional rate forever.

All of that is fair. Capacity is finite, promotions end, and a company that raises a permanent floor while retiring a temporary ceiling has not committed a scandal. If the message had simply said “the temporary boost ends on 14 September, and we’re making a smaller permanent increase in its place,” there would be little to write about beyond the ordinary grumble of a limit moving. This is not a story about a company doing something outrageous. It is a story about how a company chose to describe something ordinary.

Why the framing still grates

The pro-consumer objection is narrow and specific. When you announce a reduction by opening with an increase, you are choosing the baseline that flatters you rather than the one your users actually experience. Nobody sizing their week around Claude Code is comparing against a pre-promotion baseline they may never have seen; they are comparing against the usage bar in front of them right now. Against that bar — the only one that describes their real workflow — the number is minus 17, and that is the number that came second.

The line about “exciting changes that will make it feel like you’re getting more from Claude” is the tell. Making something feel like more is a different project from giving people more, and putting it in the same breath as a documented reduction invites exactly the reading it got. It is the same move we have flagged before, where a flat subscription quietly becomes a meter and the change is narrated as an upgrade. Here is who is affected and what it means in practice:

• Heavy Claude Code users. If you run agentic coding sessions daily and sized your week against the boosted limit, your usable weekly allowance shrinks by roughly a sixth from 14 September. Plan for it now, not on the day.
• Teams that standardised on a number. Any team that budgeted “we get X hours of agent time a week” from the current usage bar needs to redo that sum with X reduced by about 17%.
• Everyone else. If you rarely hit the limit, this changes nothing for you — which is worth saying, because not every limit change is a catastrophe, and pretending otherwise would be its own kind of hype.

The pattern this fits

There is also a quieter irony in the promise that rode alongside the cut: more “visibility and control of your usage.” Visibility is welcome, and a clearer usage meter would be a real improvement. But visibility is not headroom. Being able to watch your allowance drain more precisely is not the same as having more of it, and offering the former in the same sentence as removing some of the latter is exactly the substitution users have learned to read closely. If the promised tools arrive and are good, they will be worth having on their own terms — but they do not turn a 17% reduction into a gain, and pitching them as consolation invites the very cynicism they were presumably meant to defuse.

The reason a fairly small adjustment gets a whole piece is that it is not the first time. Claude’s allowances have been a moving target for a while — we wrote a whole explainer on why Claude’s rate limits are still confusing, and a Voices round-up on how the limits tightened and nobody sent a memo. Each individual change has a reasonable-sounding explanation. The cumulative effect is that paying users cannot form a stable expectation of what their subscription buys, because the denominator keeps moving and the announcements keep arriving pre-framed as improvements.

That is the corrosive part, and it is not unique to Anthropic. The industry has trained people to read every limit change as a possible downgrade wearing an upgrade’s clothes, which is a bad position for a company that mostly gets it right to put itself in. Anthropic’s clarification was honest; its instinct to lead with the flattering number is what will make its next genuinely good increase harder to believe.

What to do about it

Nothing here is an emergency, and 14 September is a fortnight’s notice. The useful posture is to treat today’s allowance as the promotional figure it is:

• Re-measure against the September number. Work out what your week looks like at 125, not 150, and find out before the deadline whether it still fits.
• Watch the usage bar, not the headline. The in-app meter is the honest number; the announcement is the marketed one.
• If it doesn’t fit, compare on real terms. This is the moment to weigh plans and providers against their current, non-promotional limits — the ones you will actually live with — rather than the boosted figure that is about to expire.

The limit change itself is minor and defensible. The framing is the thing worth remembering: a company told its users it was raising their limits on the day it was lowering them, and only made the second half plain after a delete and a repost. The maths works out to 17%. The trust cost of choosing the wrong number to lead with is harder to put a figure on, and it is the one Anthropic should worry about.