Claude Code Usage Limits Fall 17% Despite Permanent 25% Raise

A raise that leaves users with less

Anthropic described its September Claude Code change as more capacity. For developers who had been using the tool all summer, the result was less. On August 29, the company's ClaudeDevs account said standard weekly limits in Claude Code would rise permanently by 25% for Pro, Max, Team, and seat-based Enterprise plans starting September 14. It added that the current 50% increase would stay in place until then.1 The two numbers are measured from the same old baseline. Call the pre-promotion allowance 100: users had 150 during the promotion and dropped to 125 once the permanent level took effect.17

That works out to a reduction of roughly 16.7%, which Anthropic rounds to 17%.1029 By the company's own reckoning, the change it presented as an increase took a meaningful share of the weekly capacity paying subscribers had in early September.

How the message changed

The framing problem matters as much as the arithmetic. BleepingComputer reported that Anthropic deleted the original thread and replaced it with a clarification that stated the cut plainly: compared with the current level, the change was a 17% reduction in weekly Claude Code limits.1 Other coverage describes the same order of events. The first announcement led with the 25% figure, developers pushed back, and the follow-up acknowledged the drop from boosted levels.5

Some developers wrote what they thought the announcement should have said. Theo Browne posted an alternative draft that admitted the promotion was ending, called the result roughly a 17% decrease, and pointed to compute constraints as the reason Anthropic could keep only part of the boost.30 That critique gets at the real issue. The cut itself is defensible. Presenting it as an upgrade was the mistake.

The written record also lagged behind. One tracker found that on announcement day, Anthropic's Help Center article still said the 50% promotion would end August 31 and made no mention of the 25% figure. The page was not rewritten to describe the permanent level until early on September 14.2 The same tracker said that as of September 27, the promotion article was the only Anthropic-owned page it could find that documented the 25% level. It also reported that the general limits page and the pricing page did not mention the figure on September 14.2 For a product sold to engineering teams on monthly plans, announcements on X followed by slow documentation updates make capacity hard to plan around.

Why the summer boost raised expectations

The 50% boost was never presented as a short promotion in practice. It started May 13 and was first set to end July 13.4 Anthropic then extended it to July 19, to August 19, and to August 31, saying along the way that it hoped to make the increase permanent.36 The Help Center later moved the end date again, to September 13.2 By the time it ended, the boost had run about four months, which is long enough for teams to build their workflows around it.

That explains why the 17% cut stings. One analysis turned it into a planning rule: any workflow that uses more than about 83.3% of its old boosted weekly allowance will no longer fit unless the team changes its model mix, concurrency, scheduling, or billing route.7 The same analysis noted that Anthropic still does not publish a fixed weekly quota in tokens, hours, or messages, so teams have to track their own usage to know where they stand.7 Anthropic says usage depends on conversation length, model choice, tool calls, and effort level, so a weekly allowance does not convert into a fixed number of prompts.1

Implicator's reporting shows how this feels to users. Developers described checking their usage every half hour and rationing work across the week ahead of the change.10

Where the coverage agrees and where it differs

Coverage agrees on the core facts: 100, then 150, then 125, and the plans affected.1910 The disagreement is about which baseline to emphasize. One outlet ran Anthropic's confirmation as a headline and stressed that capacity remains 25% above the pre-May level and has not fallen below the old baseline.2 Others led with the 17% cut.110 Both readings are accurate. The more useful one for readers is the cut, because people work against the allowance they had last week, not one from before May.

Reports differ on smaller points too. The Help Center gave September 13 as the promotion's end, while the ClaudeDevs post said the 50% increase would last until September 14.9 Most reports say five-hour limits were untouched by the September change.29 One guide, however, lists a further increase to five-hour limits on September 22, which other coverage does not confirm.22 The five-hour picture is clearer from earlier in the year. On May 6, Anthropic permanently doubled Claude Code's five-hour limits for Pro, Max, Team, and seat-based Enterprise plans, and removed the peak-hour reduction for Pro and Max.2425 Those changes did not touch the weekly cap, so developers who run out of weekly capacity got no help from the May doubling.24

Paying for more use

The weekly cut makes more sense next to the tools Anthropic has built for paying past the limit. Subscribers who hit a cap can turn on usage credits, which bill at standard API rates and carry a $2,000 daily redemption limit.23 Inside Claude Code, a /usage-credits command manages that balance. Recent versions can also pause a task and resume it automatically when the limit resets.2628 Anthropic also handed out one-time cloud session credits of $100 for Pro and $250 for Max, plus occasional free limit resets.2526

The same approach showed up earlier with Fable 5. In July, Anthropic made the model a standing feature of Max and Team Premium at 50% of weekly limits. Pro and Team Standard users were moved to pay-as-you-go usage credits with a one-time $100 credit.8 Taken together, these moves suggest a strategy: included plan capacity is becoming a floor, and heavier use is pushed onto metered billing.

Code review is billed separately

That approach is most visible in Anthropic's code review product. Managed Code Review launched March 9 as a multi-agent pull-request reviewer for Team and Enterprise customers.1118 Anthropic's documentation says each review averages $15 to $25, rising with PR size and complexity, and is billed through usage credits rather than counting against plan limits.12 The trigger setting drives the bill. Reviewing after every push multiplies the cost by the number of pushes.12

The totals add up quickly. One estimate puts a 10-developer team opening 200 PRs a month at $3,000 to $5,000 in usage credits, on top of seat fees.17 Reports disagree on cost controls. One competitor claims there are no built-in spend caps.18 Other coverage describes monthly spend caps, repository-level controls, and an admin-set cap.1116 The weight of reporting favors the existence of caps, and that should be checked before relying on the competitor's claim.

Anthropic also offers cheaper review options. The local /code-review command works on any Claude Code plan and counts against normal plan usage.1617 A GitHub Action runs on the customer's own runners.17 But local reviews draw on the same five-hour and weekly limits that just shrank. That makes managed review, billed separately, the easier choice for teams with heavy workloads.

What it means for competition

Competitive pressure is part of the story. When the 50% boost launched in May, commentators widely read it as a response to OpenAI's Codex. One estimate put Claude Code's boosted Max 5x weekly cap at 45 to 60 hours of Opus, against 60 to 80 hours of Codex on ChatGPT Plus at a similar price.4 The September change gives back part of that ground.

On code review, other vendors compete on predictable pricing. CodeRabbit charges per seat, and Copilot code review is bundled into GitHub's paid tiers.1417 One benchmark cited in competitor coverage put the Claude GitHub App sixth on online F1 score, behind Copilot and cubic.17 That benchmark comes from a rival, so it deserves some skepticism.

In short, the September change is a modest permanent improvement over the old baseline, presented in a way that hid a real cut from the levels users actually had. Anthropic corrected its wording only after pushback. Its billing changes increasingly put heavy use, including automated code review, on metered credits. Teams that depend on Claude Code should plan against the 125 level, monitor their own usage, and treat future boosts as temporary unless Anthropic's documentation says otherwise.