On August 29, 2026, Anthropic’s developer account announced that starting September 14 it will permanently raise standard weekly limits in Claude Code by 25% for Pro, Max, Team and seat-based Enterprise plans — 25% measured against the standard limit that applied before the current promotion. In a follow-up post the same day, Anthropic added the sentence most coverage buried: “Compared to today, this works out to a 17% reduction in weekly limits on Claude Code.”
Both statements are accurate, and neither is a correction of the other. They measure the same change from two different starting points. The temporary boost that has been in force since May 13, 2026 lifted weekly limits 50% above the standard level; the new permanent level lands above the old standard but below where the boost has been carrying you. If you have only ever seen the 25%-above-the-old-baseline half of that sentence, your ceiling on September 14 will be lower than you expect.
A percentage change with no stated reference point is not a fact, it is a choice of framing. That is why this post indexes the three levels — 100 before the promotion, 150 today, 125 from September 14 — and shows the division. Anthropic’s 17% is arithmetically right, and once you have seen the working you can check the next announcement in about five seconds without waiting for anyone’s coverage.
Two scope notes before the detail. September 13 and September 14 are announced dates that have not yet arrived — nothing about your limits has changed as you read this. And Anthropic has published no absolute token numbers for any of these limits, before or after, so every figure here is proportional by necessity rather than by choice.
- 01One change, announced with two true percentages.Anthropic said standard weekly limits in Claude Code rise 25% on September 14 — measured against the pre-promotion standard baseline. Anthropic also said the same change is a 17% reduction measured against the boosted level available today. Neither number is wrong; the reference points differ.
- 02Index the levels and the arithmetic is one line.Call the pre-promotion standard weekly limit 100. The temporary 50%-above-baseline boost makes today 150. The announced permanent level is 125. 125 ÷ 150 = 0.833, which is 16.7% below today and rounds to the 17% Anthropic itself states.
- 03The boost is announced to run to September 13.The temporary increase that began May 13, 2026 was already scheduled to end; Anthropic’s announcement extends it to September 13 and replaces it on September 14 with a higher permanent standard level. Pro, Max, Team and seat-based Enterprise plans are the named scope.
- 04Weekly limits only — not the five-hour window.The promotion that is ending applied to weekly usage limits only, with five-hour limits and other Claude surfaces explicitly untouched, per Anthropic’s Help Center. The September 14 announcement speaks only of standard weekly limits as well.
- 05The vendor’s durable docs lagged the announcement.When we retrieved Anthropic’s own promotion page on August 30, 2026 at approximately 09:58 UTC, it still documented the boost as ending August 31 and made no mention of September 14. Plan capacity off the announcement, then re-check the support article.
- 06Anthropic’s 17% cut is not undone by a 17% rise.Going from 150 to 125 is a 16.7% reduction measured against 150. Getting back from 125 to 150 would require a 20% increase measured against 125. That is our arithmetic on Anthropic’s own three numbers, and it is why percentage direction matters in budget models.
01 — The AnnouncementWhat Anthropic actually posted on August 29. #
The announcement arrived as a short thread of three posts from Anthropic’s developer account on August 29, 2026. Because X returns 403 to unauthenticated fetches, the text below is taken from an unrolled mirror of the thread and cross-checked against two independent corroborations — a contemporaneous news report and a Hacker News commenter who quoted all three posts in full.
The first post is the headline: “Starting September 14, we’re permanently raising standard weekly limits in Claude Code by 25% for Pro, Max, Team, and seat-based Enterprise plans. Until then, the current 50% increase will be in place.” Two things to hold onto. The 25% is measured against the standard weekly limit — the level that applied before the current promotion, not the level you are using today. And the “current 50% increase” is likewise 50% above that same standard baseline.
The second post is the one that makes this a story rather than a press release:
"Compared to today, this works out to a 17% reduction in weekly limits on Claude Code. 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. Can't wait to share them."— Anthropic’s developer account, August 29, 2026
That 17% is Anthropic’s own figure, published by Anthropic, and it is correct. It is measured against the boosted level available today, not against the old standard baseline — which is exactly why it can sit beside a 25% increase without contradiction. Section 02 shows the division.
The third post is the only sentence in the thread that touches on why: “Thanks for hanging with us while we figured out what we can sustainably serve going forward. Let us know what you build this weekend!” That is the ceiling on motive. Anthropic has published no capacity analysis alongside this change, and we are not going to infer one — a vendor saying it worked out what it can sustainably serve is a statement about process, not a disclosed cost model.
One piece of the story is not ours to assert. BleepingComputer reported on August 29, 2026 that “Anthropic later deleted the original thread and posted a clarification, where it clearly admits the reduction.” We did not observe the deletion, and the deleted text is not recoverable, so that sequence belongs to BleepingComputer’s report rather than to our record. What we can verify is the surviving thread, which carries both Anthropic’s 25%-above-the-old-baseline headline and Anthropic’s 17%-against-today clarification.
The promotion now ending is one we covered when it was extended: the +50% weekly boost and Fable 5 landing in Max and Team documented the promotion’s shape and argued that a deadline which keeps moving is a pricing experiment rather than a deadline. That post owns the pattern; this one owns the number.
scheduled, not elapsed. Until then the boost remains in force at 50% above the standard baseline, per Anthropic’s own announcement.
02 — The ArithmeticIndex the levels: 100 → 150 → 125. #
Anthropic publishes no absolute token numbers for weekly limits, so the only way to reason about them is proportionally. Index the pre-promotion standard weekly limit at 100. The temporary promotion lifts it 50% above that baseline, so today is 150. The announced permanent level is 25% above the same baseline, so from September 14 it is 125. BleepingComputer laid out the same three numbers on August 29: “If Claude Code’s original weekly allowance was 100, the temporary boost gives you 150 today. On September 14, that drops to 125.”
With three indexed levels on the table, both percentages fall out of one line of arithmetic each. Against the old standard baseline: 125 ÷ 100 = 1.25, an increase of 25% — Anthropic’s headline. Against the level available today: 125 ÷ 150 = 0.8333, so the new level is 83.3% of today’s and therefore 16.7% below it, which rounds to the 17% reduction Anthropic states in its own clarification.
Where the 17% comes from · 125 ÷ 150 = 0.833
Indexed levels per Anthropic’s announcement, August 29, 2026The graphic is the whole post in one figure: the third bar is taller than the first and shorter than the second, and each of Anthropic’s two percentages describes one of those two comparisons. A reader who only knows “+25%” is reading the first comparison and will be surprised by the second.
One further piece of arithmetic, ours rather than Anthropic’s, and labelled as such because it matters for budget models: the 17% cut Anthropic describes is not undone by a 17% rise. Moving from 150 to 125 is a 16.7% reduction measured against 150. Moving back from 125 to 150 would require a 20% increase measured against 125. Percentage movements are not reversible at the same magnitude, which is a recurring source of error when a promotional level is treated as the planning baseline.
03 — Reference PointsThe two 25s are the trap. #
There are two different “25%” figures circulating about this announcement, and only one of them is right. Anthropic’s +25% — measured against the old standard baseline — is correct and is the vendor’s own framing. A “−25% cut” measured against today is wrong. The arithmetic that produces it is subtracting the two boost percentages from each other (50 − 25 = 25) instead of comparing the two levels (125 versus 150). Some coverage published the wrong version, in at least one case contradicting itself inside a single document: Anthropic’s correct 17%-against-today figure in the body, and “about 25%” as the cut against today in both the summary and the FAQ. We are not naming publications against a wrong number; the correction is the service, not the target.
This is also not a case of readers being careless. On the Hacker News thread about the announcement, the most-upvoted submission title itself described the change as a 25% cut against today’s level, and commenters corrected it in-thread within the hour. One commenter, gs17
, wrote: “I’m pretty sure they mean going from 150% to 125%, where 1.25/1.50 = 0.83, so they’re calling that a 17% reduction.” Another, giuliomagnifico
, read it the other way entirely — “‘Reduce limits’?! They’re rising the limit, not reducing” — and was answered by tomasphan
: “They are reducing to +25% from current +50%.” These are individual commenters, not a consensus, but the pair is the clearest available illustration that the two-reference-point framing confused careful readers in both directions.
The audit below is the transferable part. Each row states a version of the change in circulation, the reference point it uses, and whether it is right. Notice that the column doing the work is not “figure” but “reference point” — that is the lesson, and it applies to every promotional expiry in this market, not just this one.
| Statement in circulation | Reference point it uses | Figure | Correct? |
|---|---|---|---|
| Correct — a stated reference point you can check | |||
| “Permanently raising standard weekly limits by 25%” | Pre-promotion standard baseline (100 → 125) | +25% | Yes — Anthropic’s own headline |
| “A 17% reduction in weekly limits on Claude Code” | Today’s boosted level (150 → 125) | −17% | Yes — Anthropic’s own clarification, rounded |
| “Roughly a 16.7% reduction” | Today’s boosted level, unrounded (150 → 125) | −16.7% | Yes — the exact form of the same figure |
| “From about 150% of baseline to 125%” | Indexed to the baseline, no percentage change quoted | Levels only | Yes — and the hardest version to misread |
| Incomplete or wrong — no reference point, or the wrong operation | |||
| “Limits are going up” | Pre-promotion standard baseline, direction only | Directional | True but incomplete — silent on today’s level |
| “A cut of about 25% compared with what you have now” | Neither level — subtracts the two boost increments (50 − 25) | −25% | No — the operation is wrong, not the data |
Sources for the table: Anthropic’s August 29, 2026 announcement thread for the two correct percentages; our own arithmetic on Anthropic’s three indexed levels for the unrounded form; published coverage and the Hacker News thread for the incorrect version, which we describe without attribution by name. If you take one habit from this post, take the middle column: ask what a percentage is measured against before you write it into a budget.
04 — ScopeWhat is not changing on September 14. #
The blast radius is narrower than the coverage volume suggests, and the boundaries come from Anthropic’s own words in two places: the announcement thread, and the Help Center article on the promotion that is ending (Anthropic’s own short link; canonical article at support.claude.com/en/articles/15910845
).
Weekly limits only. The Help Center article states that “5-hour usage limits are not affected by this promotion” and, in its FAQ, that “this promotion increases weekly usage limits only.” The September 14 announcement speaks only of standard weekly limits. Claude Code’s five-hour rolling window — the short-cycle cap that Anthropic permanently doubled on May 6, 2026 — is not named in either document.
Claude Code, not Claude. The Help Center is explicit that “usage limits for other Claude products, like Claude (web, desktop, and mobile) and Claude Cowork, are unchanged” by the promotion, and the September 14 announcement names Claude Code only.
Plan scope, stated carefully. The announcement names Pro, Max, Team and seat-based Enterprise plans. The promotion that is ending excluded two groups by name — “free plans and consumption-based Enterprise seats are not included in this promotion.” The September 14 post does not mention either group. That is an absence, not an exclusion: we can say the promotion named them as out of scope, and that the new announcement is silent about them.
No pricing or tier change is mentioned. The announcement is about limits. It introduces no new subscription tier and says nothing about plan prices. If you are working out where this sits in the wider pattern of what a paid AI plan actually gives you, our reference on included versus metered AI pricing is the conceptual frame; the API-side equivalent — what happens when a budget ceiling rather than a plan ceiling is hit — is covered in our spend-cap exhaustion census.
Standard weekly limits, Claude Code
The announcement raises the standard weekly limit 25% above the pre-promotion baseline for Pro, Max, Team and seat-based Enterprise plans, and describes the same move as 17% below today’s boosted level.
The five-hour rolling window
The ending promotion applied to weekly limits only and said five-hour limits were unaffected. The September 14 announcement names weekly limits only. The five-hour window Anthropic doubled on May 6, 2026 is not mentioned.
Absolute numbers published
Anthropic has published no token figures for these weekly limits before, during or after the promotion. Every number in the announcement — and therefore in this post — is proportional.
05 — Documentation LagThe vendor’s own docs said something else. #
This is the most actionable line in the story and almost nobody led with it. Anthropic announced the September 14 schedule on its developer account. Anthropic’s durable documentation for the promotion that is ending had not caught up.
When we retrieved the promotion page through Anthropic’s own short link on August 30, 2026 at approximately 09:58 UTC, it was still titled “Claude Code May–August 2026 weekly limits promotion,” still marked “Updated over a week ago,” still carried the note that “increased weekly limits now run through August 31, 2026,” and still stated terms “valid from May 13, 2026 through August 31, 2026 at 11:59 PM PT.” It contained no mention of September 14 and no mention of the permanent increase. Its own FAQ answer for what happens next — “after August 31, 2026, weekly usage limits in Claude Code return to their standard levels” — was, by then, superseded twice over: the boost is announced to run to September 13, and the standard level itself is announced to rise.
That is a description of a live page at a moment, not a permanent condition and not a claim about what the page said on August 29. By the time you follow the link it may well be corrected, which is exactly the point: the retrieval timestamp is part of the fact. A second outlet independently observed the same gap in the same window, so this is not a single anomalous read on our side.
Anthropic’s promotion page still documented the boost as ending August 31, 2026 at 11:59 PM PT, with no mention of September 14 or of the permanent 25%-above-baseline standard limit. Retrieved via Anthropic’s own short link clau.de/cc-50-promo
.
The operational rule this produces: the fast channel and the durable channel are not the same channel. Plan capacity off the announcement, then re-check the support article before you treat the transition mechanics as final.
06 — VerificationCheck it in your own client. #
Because Anthropic publishes no token numbers, the only usage figures with your name on them are the ones your own client reports. Two surfaces are worth knowing.
The /usage
command in Claude Code shows your current weekly consumption, including a “Current week (all models)” figure. And Claude Code’s status-line JSON exposes machine-readable rate-limit keys, which is useful if you want to watch the numbers over the cutover rather than eyeball them. A GitHub issue opened on August 20, 2026 documents the keys present in build 2.1.237 — rate_limits.five_hour
, rate_limits.seven_day
, rate_limits.seven_day_opus
and rate_limits.seven_day_sonnet
— each carrying a used_percentage
and a resets_at
. Note the shape that implies: the weekly allowance is a shared pool with model-scoped ceilings inside it, and the same issue notes that /usage
already shows a separate “Current week (Fable)” limit alongside “Current week (all models).”
A caveat on what a weekly number means. As BleepingComputer summarised it on August 29, Anthropic says Claude usage varies with conversation length, model choice, tool usage and effort level, so a weekly allowance does not translate into a fixed number of prompts. A 17% reduction against today’s level — Anthropic’s figure — is a 17% reduction in a pool whose exchange rate into actual work you set yourself, largely through model choice. If your ceiling is about to be lower, our guide to which Claude model to use when is the practical lever, since routing the cheap steps away from your most expensive model buys back headroom without any vendor action.
Today: 150
. From September 14: 125
. Then 125 ÷ 150 = 0.833
, so the announced level is 16.7% below today’s — the 17% reduction Anthropic states. Against the pre-promotion baseline of 100
, the same 125 is a 25% increase, which is Anthropic’s headline.
In your client: run /usage
for “Current week (all models)”, or read `rate_limits.seven_day.used_percentage`
from the status-line JSON if you want to log the cutover.
07 — SequenceWhere September 14 sits in the run of changes. #
This is a strip, not a ledger — just enough dated points to place the cutover in sequence. Every entry below is a Claude Code weekly-limit event; the separate lineage of Fable 5 plan access is a different sequence and is deliberately not merged in here.
| Date | Event | Status |
|---|---|---|
| May 6, 2026 | Anthropic permanently doubles Claude Code’s five-hour rate limits and removes the peak-hours reduction for Pro and Max. Weekly limits unchanged. | Elapsed · first-party post |
| May 13, 2026 | The temporary weekly promotion begins: weekly limits 50% above the standard baseline, first scheduled to expire July 13. | Elapsed · Help Center confirms |
| Jul–Aug 2026 | The promotion’s end date moves three times before this announcement, landing on August 31. We do not assert the announcement channel for the intermediate moves. | Elapsed · dates only |
| Aug 29, 2026 | Anthropic announces the September 14 permanent standard limit at 25% above baseline, and clarifies the same change as a 17% reduction against today’s level. | Elapsed · the announcement |
| Sep 13, 2026 | Announced last day of the temporary boost — two weeks past the August 31 date the Help Center still documented at our retrieval. | Announced · not yet elapsed |
| Sep 14, 2026 | Announced first day of the higher permanent standard weekly limit: 125 on the indexed scale, up from a baseline of 100 and down from today’s 150. | Announced · not yet elapsed |
Counted on the weekly-limit lineage alone, the September 13/14 reset is the fourth change to that promotion’s end date since it began in May. Counted as “deadlines that moved,” the pattern is older than this promotion and wider than Anthropic — which is why the cross-vendor picture lives in a separate asset rather than in this post. Our ledger of every AI coding plan limit change in 2026 is where the full dated table sits, with each row’s source and each vendor’s own framing recorded next to the effective direction. For the argument that a repeatedly moved deadline should never be a planning baseline, our earlier post on access whiplash from repeated extensions made that case in July; this announcement is the strongest evidence for it so far, in both directions.
08 — ActionWhat to do with the two weeks before September 14. #
First, work out whether you are actually exposed. If your team has never hit a weekly ceiling, a move from 150 to 125 on the indexed scale changes nothing you will notice — you are still above the 100 baseline you were budgeted against before May 13. The teams that will feel September 14 are the ones whose weekly consumption has grown into the boost, because they have been planning against a level that was always documented as temporary.
Second, measure now, while the boost is still on. Log your weekly usage for the next fortnight from /usage
or the status-line keys. A percentage of a pool you have never measured is not a number you can plan with, and the comparison you will want in mid-September is against your own pre-cutover consumption, not against a vendor percentage.
Third, budget against the announced permanent level, not the boost. The durable planning number from September 14 is 125 on the indexed scale — 25% above the pre-promotion baseline. Treat any future extension or further increase as upside rather than as an assumption. If seat economics are part of the decision, the inputs change when the ceiling does: our Claude Code ROI formulas for teams take the ceiling as a variable, so re-running them with 125 instead of 150 is a ten-minute exercise rather than a rebuild. The wider pricing-model context — how subscription, credit and metered plans differ in exactly this failure mode — sits in our comparison of the four ways AI coding is priced now.
Fourth, calendar a documentation re-check. Anthropic announced the change on its developer account while its own support article still described the previous schedule at our retrieval on August 30. Put a reminder in for early September to re-read the support article, and treat the announcement as the operative source until the durable documentation agrees with it.
Absorb the change
You have never hit a weekly ceiling. From September 14 your standard limit is 25% above the baseline you were on before May 13, so the cutover is invisible. Log a week of /usage anyway so the next announcement finds you with a measurement.
Route work cheaper
You occasionally hit the weekly ceiling under the boost. Because the weekly pool is shared across models with model-scoped ceilings inside it, moving routine steps off your most expensive model buys back headroom without any change to your plan.
Re-budget the seats
Your team plans capacity against the boosted level. Re-run seat maths at 125 on the indexed scale rather than 150, decide where the overflow goes — more seats, API fallback, or narrower agent scope — and write down which assumption you chose.
09 — ConclusionThe reference point is the fact. #
Two true percentages, one checkable change.
Anthropic announced one change on August 29, 2026 and described it with two percentages. Standard weekly limits in Claude Code rise 25% above the pre-promotion baseline on September 14. The same move is, in Anthropic’s own words, “a 17% reduction” against the boosted level available today. Index the three levels — 100, 150, 125 — and both statements resolve in one line of arithmetic each, with Anthropic’s 17% coming out correct.
The generalisable lesson is smaller than the news and lasts longer. A percentage change with no stated reference point cannot be checked, and in this market the same change is routinely quoted against whichever baseline flatters the framing. Coverage of this announcement carried at least six different versions of the number, two of them arriving at a wrong figure from correct data purely by picking the wrong comparison. The defence is not trusting a better outlet; it is asking what the percentage is measured against, every single time.
One practical caution beyond the arithmetic. On the day of the announcement the vendor’s fast channel and the vendor’s durable documentation disagreed about when the boost ends, and at our retrieval on August 30 the support article still described the older schedule. If you plan capacity off support documentation, build in a re-check rather than assuming the page is current. Measure your own weekly consumption while the boost is still running, budget against 125 rather than 150, and September 14 lands as a configuration fact rather than a surprise.