Today, September 14, Anthropic permanently raised Claude Code’s standard weekly usage limits by 25% across Pro, Max, Team, and seat-based Enterprise plans. That’s the headline. It’s also, depending on how you were using Claude Code yesterday, a cut. A 17% one. Anthropic had to say so itself, in a deleted-and-reposted clarification, after developers spent a week pointing out the math didn’t add up.

Here’s the part that got buried in the original post: Anthropic had been running a temporary 50% boost to those same weekly limits since May. That boost expired yesterday, September 13. The new “permanent increase” is 25% over the original baseline — not 25% over what anyone’s actually been working with for the past four months. Go from a 50% boost to a 25% boost and you’ve lost ground, not gained it.

Quick Summary: What Changed

Detail	Info
Effective	September 14, 2026
Original announcement	Anthropic: permanent 25% increase to Claude Code weekly limits
What expired	A temporary 50% weekly-limit boost running since May 2026, ended September 13
Net effect	A 17% cut versus the limits users had access to through September 13
Who’s affected	Claude Code Pro, Max, Team, and seat-based Enterprise plans
Anthropic’s clarification	”Compared to today, this works out to a 17% reduction in weekly limits on Claude Code.”

Bottom line: Anthropic marketed today’s change as a permanent increase. After backlash, it admitted the change users will actually feel is a 17% cut.

What Actually Happened

Do the arithmetic the way Bleeping Computer laid it out: say the original May baseline was 100 units of weekly Claude Code capacity. The temporary boost took that to 150. Users built their workflows around 150. Today, that number drops to 125 — 25% above the original 100, but 17% below the 150 everyone’s actually been running on since May.

Anthropic’s own original post on X read: “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.” Technically accurate. Also structured so the number you see first is the one that sounds good, and the number that determines your actual week-to-week capacity — the expiring 50% boost — reads like a footnote.

Developers didn’t let it stay a footnote. The Decoder reported that a community note landed on the original post within hours, spelling out the 150-to-125 math in plain terms. Theo (t3.gg) put it bluntly on X: “Anthropic increased Claude Code rate limits by -17%! Wait that’s not right…” Enough pressure built that Anthropic deleted the original thread entirely and reposted with an actual admission attached.

What Did Anthropic’s Clarification Actually Say?

1. It confirmed the cut in plain language — “Compared to today, this works out to a 17% reduction in weekly limits on Claude Code,” the reposted thread reads, a sentence absent from the original announcement.
2. It reframed the increase as relative to the pre-boost baseline — the 25% figure was always measured against May’s original limits, not against the 50% promotional level users had been on since.
3. It promised future changes, without specifics — “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,” per the same clarification thread.
4. It offered no compensation or transition period — unlike Anthropic’s April cutoff of third-party Claude tool access, which came with a one-time credit, this change shipped with nothing beyond the promise of “exciting changes” to come.

Why This Matters

Anthropic didn’t lie. Every number in the original post was true. That’s what makes this a communications failure rather than a factual one — and also exactly the kind of failure that’s hard to walk back once people notice, because the defense (“we said 25% and we meant 25%”) is technically airtight and still leaves users feeling like they got played. Notebookcheck’s framing — a 25% increase with a 17% catch — is the more honest headline than the one Anthropic shipped.

This isn’t Anthropic’s first run through this exact playbook this year. Back in April, Anthropic quietly cut Claude’s default reasoning effort for five weeks without disclosure, got caught by a forensic GitHub audit from an AMD director, and reversed course only after Fortune covered the backlash. The pattern rhymes: a change that saves Anthropic compute, communicated in a way that undersells the downside, followed by public pressure forcing a fuller accounting. The company that markets itself as the safety-and-transparency alternative to OpenAI keeps handing critics evidence for the opposite claim.

It also lands one day after we covered Dario Amodei’s essay arguing the whole industry needs to slow down and submit to more external scrutiny. That essay was about frontier model safety, not usage limits — different subject entirely. But a company asking the industry to trust it with more transparency around dangerous capabilities looks worse doing damage control on a usage-limit tweet the very next week, even on an unrelated issue. Optics compound.

The scale here matters too. This isn’t a niche API change. It hits every Claude Code Pro, Max, Team, and seat-based Enterprise account — the same tiers this site covers every time Anthropic touches permission defaults or execution environments. For teams that scaled up agentic workflows during the May-to-September boost window, a 17% cut isn’t cosmetic. It’s the difference between finishing a sprint’s worth of agent sessions inside your weekly allotment and hitting a wall on Thursday.

What Are Your Options Now

Audit your last four months of usage before you hit a wall. If your team’s Claude Code usage grew to fit the 50% boost, going back to 125% of baseline could mean throttling mid-week. Pull your usage data now, not after you get rate-limited on a deadline.

Don’t assume Enterprise insulates you. Seat-based Enterprise plans are explicitly included in this change. If your org negotiated Claude Code access as part of a broader Anthropic contract, confirm with your account rep whether your specific terms follow the standard weekly-limit schedule or a custom one.

Watch for the “exciting changes” Anthropic teased. The clarification thread promised more visibility and control over usage without a timeline or specifics. Until that ships, treat it as a future consideration, not a current mitigation.

Budget for API overflow. If weekly limits become a recurring bottleneck, pairing your Pro, Max, or Team plan with pay-per-token API access for overflow work is the same fallback Anthropic pointed third-party tool users toward earlier this year. It costs more per token, but it doesn’t run out on a weekly clock.

The Bigger Picture

Temporary promotional boosts that quietly expire into a smaller “permanent increase” aren’t unique to Anthropic — it’s a familiar move across subscription software generally. What’s unusual is doing it to a developer audience that reads changelogs and does arithmetic in public, on a platform, X, built specifically for someone to reply with the exact math that undercuts your framing within the hour.

Anthropic’s core problem this year hasn’t been making defensible resource-management decisions. Reducing default reasoning effort to manage compute cost was defensible. Tightening weekly limits back toward baseline after a promotional period ends is defensible too — promotions end, that’s what makes them promotions. The problem, both times, has been leading with the framing that makes Anthropic look best and letting users discover the fuller picture themselves. Do that twice in one year to the same audience and the benefit of the doubt runs out.

Our Take

We think the underlying decision here is fine and the messaging around it wasn’t. Promotional boosts expiring is normal. Anthropic didn’t owe anyone a permanent 50% bump forever. What it owed users was a first sentence that said “your limits are dropping 17% from what you’ve had since May” instead of one that led with a bigger-sounding number measured against a baseline most active users hadn’t touched in months.

The “exciting changes” line in the clarification is doing a lot of work for a promise with zero specifics attached. Anthropic has now told users, in successive incidents this year, that it’s got improvements coming for effort defaults, for pricing, and now for usage visibility. Believe it when it ships. Until then, plan your team’s Claude Code usage around the number you actually have this week, not the one Anthropic would prefer you focus on.

Frequently Asked Questions

Is Claude Code’s weekly limit actually going up or down on September 14?

Both, depending on the comparison point. It’s up 25% versus Claude Code’s original May 2026 baseline. It’s down 17% versus the temporary 50% boost that had been active since May and expired September 13. For anyone who’s been using Claude Code during that boost period — which is most active users — the change you’ll feel is the 17% cut.

Which Claude Code plans are affected by the weekly limit change?

Pro, Max, Team, and seat-based Enterprise plans. Anthropic’s announcement names all four tiers explicitly, so this isn’t limited to individual subscribers — teams and enterprise accounts on seat-based billing are included.

Why did Anthropic delete its original announcement?

Anthropic’s original post led with the 25% increase and didn’t clearly state that current users would see a net reduction. After a community note and public pushback from developers, Anthropic deleted the thread and reposted a clarification that directly states: “Compared to today, this works out to a 17% reduction in weekly limits on Claude Code.”

What was the temporary 50% boost, and why did it end?

Anthropic had been giving Claude Code users 50% more weekly usage capacity than the original baseline since May 2026. Anthropic hasn’t published a detailed reason for letting the promotion expire rather than extending it; the September 14 change replaces it with a smaller, permanent 25% increase over the original baseline instead.

Did Anthropic offer any compensation for the reduction?

No. Anthropic’s clarification promised unspecified future changes — “more visibility and control of your usage” — without a timeline, dollar figure, or compensating credit. That’s a different response than Anthropic’s April 2026 cutoff of third-party Claude tool access, which came with a one-time credit and a discount on prepaid usage bundles.

How does this compare to Anthropic’s other 2026 transparency issues?

It follows a similar shape to Anthropic’s undisclosed reduction in Claude’s default reasoning effort earlier in 2026, which ran for five weeks before a forensic audit and press coverage forced a public reversal. In both cases, a resource-management decision that Anthropic could have defended openly instead surfaced through user detection and public pressure, which made the transparency failure the bigger story than the underlying change.

What should heavy Claude Code users do right now?

Check your usage over the past four months against the new 125%-of-baseline limit, not the 150%-of-baseline limit you’ve had access to since May. If your workflows were sized for the boosted level, budget for either hitting weekly caps sooner or supplementing with pay-per-token API access for overflow work.

---

Last updated: September 14, 2026. Sources: Anthropic — original Claude Code weekly limits announcement · Anthropic — clarification admitting the 17% reduction · Bleeping Computer — Anthropic is cutting Claude Code’s current weekly limits by 17% · The Decoder — Anthropic’s Claude Code limit change is a raise on paper but a cut in practice · Notebookcheck — Anthropic announces a 25% increase to Claude Code limits, but there’s a 17% catch · Theo (t3.gg) on X.

Related reading: Claude’s Hidden Performance Cut: What Users Found · Anthropic Cuts Third-Party Claude Tool Access · Claude Code Just Stopped Asking Permission · Claude Code Self-Hosted: The Fix for Rogue Agents? · Dario Amodei Told AI Labs to Slow Down. OpenAI Blinked.