Anthropic's 'permanent 25% increase' to Claude Code limits is actually a 17% cut from what you have now
The temporary 50% boost to Claude Code weekly limits ends 13 September 2026. The permanent replacement is 25% above the original baseline: not today's.
Anthropic announced on 29 August 2026 that it would permanently raise Claude Code’s standard weekly usage limits by 25% above the original pre-promotion baseline, starting 14 September 2026. That sounds like good news. The catch is in what it replaces.
Since 13 May 2026, users on Pro, Max, Team, and seat-based Enterprise plans have been running on a temporary 50% boost to those weekly limits. Anthropic extended that boost repeatedly across the summer, first through mid-July, then through 19 July, then 19 August, then 31 August, and finally through 13 September 2026. At 11:59 PM PT on 13 September, the temporary boost expires. On 14 September, the new “permanent” level kicks in.
If the original baseline is 100 units, users have been getting 150. From 14 September, they get 125. That is a 16.7% reduction from what users have had all summer. Anthropic acknowledged this directly in its own follow-up communications, describing the change as “a 17% reduction” relative to the level available at the time of the announcement.
What’s actually changing
The framing of a “permanent 25% increase” is technically accurate measured against the original May baseline. It is also genuinely better than reverting entirely to that baseline, which was the alternative each time the temporary boost was extended. But if you have been working inside Claude Code since May, your practical experience from 14 September will be a smaller weekly allowance than the one you have had for the past four months.
A few specifics worth knowing:
Only the weekly cap is changing. The doubled 5-hour session limits introduced in May are not going away. Those stay. The rolling 5-hour window you use to pace work within a session is unaffected.
The change applies to Pro, Max, Team, and seat-based Enterprise plans. Free plans and consumption-based Enterprise seats are not affected by this change.
Usage pools across products. Claude Code, Claude.ai chat, and Cowork all draw from the same usage pool. If you are heavy across more than one of these, a tighter weekly cap will register faster.
You can check your actual limit by running /usage in the Claude Code CLI. Anthropic publishes multipliers rather than absolute token counts, so the only way to know your specific ceiling is to check directly.
A few things Anthropic is adding alongside the cut
The Claude Code changelog for week 33 includes two additions that soften the impact at the edges.
Auto-continue. Claude Code version 2.1.234 added automatic continuation when a session or usage limit resets, available in both the CLI and the desktop app. Previously, hitting a limit meant your session stalled and waited for a manual restart. With auto-continue enabled (it is on by default, configurable via /config), the session picks back up automatically when the reset window opens. In the desktop app, the usage limit card shows the retry time and includes a checkbox to control the behaviour. Note that the auto-continue option appears on the session-limit card but not on the weekly-limit card.
The auto mode classifier no longer burns quota. Auto mode uses a background permission classifier to approve or block actions without prompting you each time. That classifier was previously counting against your usage limits. It no longer does. Auto mode became the default permission mode for new sessions on Pro, Max, and Team plans in mid-August 2026, so this is a meaningful fix for anyone who adopted it quickly.
What this means for you
If you rarely bump against the weekly cap, 14 September will pass without much notice. The new limit is still 25% above where things started in May, and the session limits are unchanged.
If you are a heavy Claude Code user who has been pushing close to the ceiling by the end of the week, this is worth planning for now rather than discovering mid-sprint in late September.
A few practical adjustments:
Front-load the heavy work. If your week resets on a fixed day, schedule the most intensive agentic tasks earlier in the week rather than letting them accumulate toward the end when you may have less headroom.
Be deliberate about model choice. Sonnet is the default and handles the large majority of coding tasks well. Opus offers deeper reasoning for genuinely hard problems, large cross-cutting refactors, difficult debugging, architectural decisions, but it draws more from your quota. Saving Opus for situations where it actually changes the output matters more when the weekly cap is tighter.
Usage credits remain an option. If the new limits genuinely are not enough for a particular week, Anthropic allows you to purchase usage credits beyond your plan cap at API rates. It is not a substitute for adequate limits, but it is a valve for high-demand weeks rather than a hard stop.
The broader picture
Anthropic’s July communications hedged explicitly: it wanted to make part of the summer boost permanent but warned that compute capacity remained tight. September 14 is that partial retention arriving. Half the promotional increase, made permanent. The context matters because the 50% boost was never framed as a baseline shift, even if four months of usage at that level has made it feel like one.
The compute that powered the May expansion came from a significant infrastructure deal, and Anthropic has been managing demand against that capacity ceiling throughout the summer. A permanent 25% raise above the original baseline is a real commitment. It is just not the commitment that users who joined or ramped up usage in May may have been expecting.