Claude Code's Weekly Limits Change on September 14 - a Permanent 25% Raise That Sits About 17% Below Today

The change to Claude Code's weekly usage limits that Anthropic announced in late August takes effect on September 14, 2026. It is a permanent 25% raise against the original baseline, but the temporary 50% boost ends the same day, leaving about 17% less than the level just before.

Claude-Code Anthropic developer-tools pricing rate-limits AI-coding-tools

Claude Code’s weekly usage cap moves to a new level on September 14, 2026. In late August, according to reports, Anthropic posted on its official X account: “Starting September 14, we’re permanently raising standard weekly limits in Claude Code by 25% for Pro, Max, Team, and seat-based Enterprise plans”¹.

The baseline that the raise is measured against, however, is the allowance from before the temporary measure. A temporary 50% boost was in force when the announcement went out, and it ends on the same day. Anthropic later issued a correction acknowledging that, against the current level, this amounts to a reduction of roughly 17%, as reported¹.

The Same Change Reads as a Raise or a Cut Depending on the Baseline

Both “25% more” and “about 17% less” describe the same change because they are measured against different reference points. In the worked example carried by reporting, an original weekly allowance of 100 becomes 150 while the temporary boost applies, and 125 from September 14¹. Against 100 that is a 25% increase; against the 150 in force just before, it is about 17% lower.

Anthropic deleted its initial post and published a correction stating that, compared with today, the weekly limits on Claude Code fall by 17%, according to the report¹. The Decoder characterises the change as an increase on paper and a cut in practice².

For anyone keeping the same workload, the point in the week at which the allowance runs out arrives earlier than before.

Two Layers of Limits, and a Pool That Claude Code Shares

What changes here is the weekly layer, but Claude’s usage limits have always had two. Anthropic’s pricing page describes a five-hour window that rolls forward and resets on every plan, with paid plans stacking a weekly cap above it³.

That allowance is not reserved for Claude Code. As the same page puts it, one shared pool feeds Claude on the web, on desktop, on mobile and inside Claude Code alike³. Time spent chatting with Claude outside the editor eats into the same weekly figure.

How many prompts the allowance corresponds to is not something Anthropic publishes. By its own account, what a plan yields shifts with how long and how involved a conversation gets, which model is picked and which features are switched on, leaving no fixed count of messages³. Reporting adds the same caveat: consumption moves with those factors as well as with tool use, so the weekly figure cannot be translated into a set number of Claude Code prompts¹. The concrete numbers behind each plan’s weekly limit - hours or tokens - have not been disclosed.

The reason consumption climbs the longer a session carries a large context is that both billing and limits are built on the token as a unit. The /skill-doctor command added in early September, which surfaces loaded skills that go unused along with the context they consume, assumes the same arithmetic (our article on Claude Code 2.1.261).

What to Do When the Allowance Runs Out

Anthropic’s pricing page sets out three routes³: sit out the reset, step up to a higher plan, or - if the plan is a paid one - enable usage credits and carry on at standard API rates.

The third of these switches from the subscription allowance to metered billing, and can be read as the fallback for workloads that exhaust the cap late in the week. Past that point, though, API unit prices apply directly. Whichever route is taken, the choice is hard to make without first knowing which models and which tasks are absorbing the consumption.

The pricing page also states plainly that Anthropic may impose weekly and monthly caps, or restrict usage by model or feature, at its own discretion in order to manage capacity and keep access fair across users³. That a revision like this one can happen is set out there in advance.

In May, the Five-Hour Limits Were Doubled

The other layer, the five-hour limit, moved in the loosening direction on May 6, 2026. Anthropic said then that the five-hour rate limits in Claude Code were being doubled across the Pro, Max, Team and seat-based Enterprise plans, and that the peak-hours reduction was being dropped for Pro and Max accounts. The same announcement covered an agreement to take the entire compute capacity of SpaceX’s Colossus 1 data center, which the company said would give it access within the month to over 300 megawatts of fresh capacity, amounting to more than 220,000 NVIDIA GPUs⁴.

In that announcement, Anthropic attributed the higher limits for Claude Code and the Claude API to the SpaceX partnership together with its other recent compute deals⁴. What that May announcement covered was the five-hour layer; what moves now is the weekly one. When the temporary 50% boost began is a date neither Anthropic’s own pages nor the reporting gives. Claude Code itself has kept moving in the meantime, through changes such as auto mode becoming the default and the fix for the macOS sandbox escape, on both the feature and the safety side.

In the correction, Anthropic also said it was working on changes meant to make users feel they are getting more from Claude while giving them more visibility and control over their usage, according to reports¹². What those changes contain has not been described in the reporting so far. For now, the practical response is to watch how fast the weekly allowance is being spent and decide where in the week the heaviest work belongs.