On August 30 Japan time, Anthropic announced it would permanently raise Claude Code's standard weekly limit by 25% starting September 14. Yet this is not an increase in what current users can actually use. Because the temporary 50% boost that has been in effect since May ends at the same time, the weekly limit after September 14 will actually be 17% lower than it is now. Anthropic itself acknowledged this decrease in a follow-up post. The reason "increase" and "decrease" coexist is that the two percentages use different baselines.
100 → 150 → 125: two percentages, two different baselines
If you set the pre-boost standard weekly limit at 100, this change becomes clear at a glance. With the temporary 50% boost that began on May 13, 2026, the available limit rose to 150. What becomes permanent starting September 14 is 125 — the old standard of 100 plus 25%.
| Period | Weekly limit (old standard = 100) | Status |
|---|---|---|
| Through May 12 | 100 | Pre-boost standard |
| May 13 – September 13 | 150 | Temporary 50% boost |
| From September 14 | 125 | Permanent 25% increase |
"25% increase" compares 100 to 125; "17% decrease" compares 150 to 125. Doing the math, (125−150)÷150 comes out to about −16.666...%. Rounded, that matches the 17% decrease from current levels that Anthropic cited. Users will lose 25 points from their current limit — but that's one-sixth of the current 150, not one-quarter.
So September 14 is simultaneously the day the usage cap goes up and the day the capacity users currently rely on gets cut. While making the limit permanent removes the uncertainty of an expiration date, for users who have built their workflows around the temporary limit, it means less breathing room later in the week. Note that Anthropic has not disclosed the absolute token or prompt counts for each plan. The figures 100, 150, and 125 are index numbers for comparison, not actual usage counts.
Official help page still says August 31
As of the early hours of August 31, Anthropic's standing Help Center and its later official post don't agree. The Help Center states that the temporary 50% boost runs from May 13 through 11:59 PM Pacific Time on August 31, after which the limit reverts to the standard weekly cap. In contrast, the official developer account ClaudeDevs announced that the 50% boost would continue until the September 14 switchover, at which point it would transition into a permanent 25% increase.
| Source | End of 50% boost | State after boost ends |
|---|---|---|
| Anthropic Help Center | 11:59 PM Pacific Time, August 31 | Reverts to standard limit |
| ClaudeDevs' later post | Switches on September 14 | Permanent 25% increase over old standard |
For planning purposes, the later ClaudeDevs announcement represents the current policy. However, no exact switchover time or region-by-region rollout has been specified. The fact remains that the standing page you'd check for contract terms is out of date. If the Help Center is updated before September 14, users should recheck the wording for eligible plans and the exact switchover time.
The deadline has already been extended several times, but the shift to a permanent 125 is different in nature from past extensions. Rather than pushing back the 150 limit once again, Anthropic is folding 25 of the 50 temporarily-added points into the standard limit while letting the remaining 25 points expire. For users, this means the future floor will be higher than the old standard, but the capacity they got used to over the summer won't be preserved.
What's changing: Claude Code's weekly limit
The September 14 announcement specifically names the Claude Code weekly limit for the individual Pro and Max plans, as well as the organizational Team and seat-based Enterprise plans. The official Help Center entry for the temporary 50% boost likewise covers Claude Code across the CLI, IDE extensions, desktop, and web. The usage limits for Claude's web, desktop, and mobile apps, as well as Claude Cowork, are unaffected by this promotion.
The 5-hour limit is a separate matter entirely. On May 6, Anthropic doubled the Claude Code 5-hour limit for the individual Pro and Max plans and the organizational Team and seat-based Enterprise plans. For Pro and Max, it also eliminated peak-hour reductions. The weekly 50% boost began a week later, on May 13. Lumping these two measures together — despite their different start dates and time windows — as simply "the Claude Code limit" risks misunderstanding exactly what is being reduced on September 14.
The temporary weekly boost excluded the Free plan and the usage-based Enterprise plan. Meanwhile, the new September 14 announcement lists only the four eligible plan types and says nothing about the status of the previously excluded usage types. The earlier exclusion conditions cannot simply be extended to apply to the new permanent limit.
This distinction matters in practice. Users who front-load heavy work into short bursts and hit the 5-hour limit first will still have to wait even if weekly capacity remains. Conversely, users who spread work evenly across 5-hour windows will still be stopped once they exhaust their weekly limit. What shrinks on September 14 is the latter kind of headroom — this is not an announcement that further tightens the short-term cap.
Why a 17% cut doesn't translate directly into 17% fewer tasks
Even though the weekly limit is shrinking by 17%, that doesn't necessarily mean 17% fewer completed tasks. According to Anthropic's Help Center, usage varies with conversation length and complexity, the model chosen, and the amount of reasoning effort applied. Features used — such as web search or Model Context Protocol (MCP) connectors — also affect consumption.
In Claude Code, prior conversation history, project context, and new instructions are all sent to the model with each turn. During a long debugging session involving many files and tool outputs, a later turn can end up far heavier than one near the start. The official guidance to use /clear to split off separate work into a new conversation, and /compact to compress history when continuation is needed, is also a way to get more completed work out of the same weekly budget.
Because of this, evaluating the impact of the limit change requires tracking usage rate and output under matching conditions. Claude Code's /usage command displays the plan's limit and the current rate-limit status. By comparing how many points of the weekly limit a similarly-sized task consumes before and after September 14 — using the same model and effort level — users can determine how the 17% reduction actually plays out in their own development workflow.
Since the absolute limits remain undisclosed, there's no way to externally calculate a subscription plan's value as "tokens per week." While the remaining percentage is comparable, the amount of work that percentage translates into doesn't line up across runs that use different models or features. What matters after the change isn't the nominal multiplier on the limit, but a real measurement of how many representative tasks you can actually complete.
Doubled 5-hour limits and an invisible weekly budget
Anthropic tied the May 6 doubling of the 5-hour limit to expanded compute resources. It said it had signed a contract to use over 300 megawatts and more than 220,000 NVIDIA GPUs from SpaceX's Colossus 1 data center within a month, directly improving available capacity for Pro and Max. A week later, the temporary 50% boost to the weekly limit began.
However, there's no stated explanation linking that compute contract directly to the current 17% decrease. What can be confirmed is that Anthropic has been adjusting the 5-hour and weekly limits — two caps on different time scales — independently. Which one a user exhausts first determines both how long they wait and how much work they can fit into the remaining week.
Anthropic has previewed changes intended to make usage more visible and give users more control. Since /usage already exists, it's unclear whether the new feature will offer more granular remaining-usage displays, break down consumption by model or task, or change how the limits themselves work. The phrase "feel like you're getting more" is also not a promise to increase the weekly limit again.
The value users get after September 14 won't be determined by the index number 125 alone. It will depend on whether the standing Help Center is updated to reflect the new terms, whether users can track consumption by model and task, and whether they can complete more work out of the same weekly budget. Only once those questions are answered can anyone judge whether the permanent 25% increase truly offsets the loss of the temporary boost.