How-to
Claude Code Limits After 14 September — the +25% Maths, Opus 5.5 Limits, and How to Forecast Seats
Anthropic called it a permanent 25% raise; its own follow-up called it a 17% cut. Both are true. Here is the arithmetic, which users now hit the weekly wall, and how to forecast Claude Code seats from here.
Checked on 26 September 2026 against Anthropic's announcements and the other sources at the end. Anthropic has not published per-plan token figures for the new weekly limits, so this guide works in relative terms.
On 29 August 2026 Anthropic's developer account posted two messages in one thread. The first said that from 14 September, standard weekly limits in Claude Code would rise 25%, permanently, for Pro, Max, Team, and seat-based Enterprise plans. The second said: "Compared to today, this works out to a 17% reduction in weekly limits on Claude Code."
Both statements are true. They measure from different baselines. If your team sized seats during the summer, the second number is the one that matters.
The arithmetic
Call the old standard weekly limit 100.
- 13 May – 13 September: a temporary +50% boost made the effective limit 150. It was announced to run to 13 July. It was then extended to 19 July, to 31 August, and finally to 13 September.
- From 14 September: the permanent level is 125, which is +25% on the old standard.
- Change for anyone used to the boost: 125 ÷ 150 = 0.833, a 17% reduction in weekly capacity.
After sixteen weeks, most heavy users had calibrated to 150. That is why the change feels like a cut, even though the permanent limit is higher than it was in spring.
Who hits the wall now
The useful question is not "how big is the limit", which Anthropic has not stated in tokens. It is "how close to the old limit was each developer running?" Every developer's utilisation of the weekly cap rises by a factor of 150 ÷ 125 = 1.2 for the same work:
| Share of the boosted weekly cap used before 14 Sep | Same work against the new cap |
|---|---|
| 50% | 60% |
| 70% | 84% |
| 80% | 96% |
| 83% | ~100% (the new wall) |
| 90% | 108% (blocked before the week ends) |
| 100% | 120% |
Anyone who regularly used more than about 83% of their boosted weekly allowance will now run out before the week resets, unless their work changes. Everyone below that line keeps working as before, just with less headroom.
What Opus 5.5 changed on 22 September
Claude Opus 5.5 did not change the weekly envelope, but it changed three things around it:
- Higher five-hour limits. Anthropic raised five-hour usage limits on Pro, Max, Team, and seat-based Enterprise. That helps with bursty days. It does not raise the weekly ceiling.
- A saveable rate-limit reset. Subscribers got a rate-limit reset they can save and use when they choose. Treat it as a one-off buffer for a crunch week, not as capacity to plan around.
- Fewer tokens per task. Anthropic says Opus 5.5 uses fewer tokens per task than Opus 5 and costs about 40% less on typical workloads via the API. If subscription limits track usage in a similar way, the same work on Opus 5.5 may use less of the weekly envelope than it did on Opus 5. Anthropic has not stated this for subscriptions, so measure it rather than assume it.
Separately, Auto mode's safety classifier is free on Pro, Max and Team, and since v2.1.278 on most Enterprise and API sessions too. See Claude Code Auto mode default for the gateway caveat.
How to forecast seats from here
Step 1: measure a clean fortnight. Use the two weeks from 14 September onwards, not summer weeks. For each developer, note how often they hit the weekly cap and on which day.
Step 2: sort developers into three groups.
- Comfortable (never near the cap): no change needed.
- Tight (hit the cap in the last day or two of the week): try working habits first. Move routine tasks to Claude Sonnet 5, keep Opus for hard work, and save the rate-limit reset for release weeks.
- Blocked (hit the cap mid-week, regularly): they need more capacity, whether a higher plan tier or API billing.
Step 3: price the blocked group both ways.
- Higher subscription tier: predictable, and still capped.
- API usage for overflow: Opus 5.5 is $4 input / $20 output per million tokens, with $0.20 cache reads. Claude Code's long sessions re-read a lot of context, so the cache price matters more than the headline rate. Put a spending limit on it before anyone starts.
Step 4: re-check in November. Anthropic has said it is working on changes to give users "more visibility and control" of usage, and Claude Sonnet 5.5 is due "in the coming weeks." Either could move this maths. Re-run the fortnight measurement after both land.
Common mistakes
- Forecasting from summer usage. Anything measured before 14 September assumed a 150 baseline.
- Reading the +25% headline as more capacity than August. It is 17% less.
- Buying seats to fix a model-choice problem. A developer running every small task on the most expensive model burns the envelope fastest. Route first, then buy.
- Treating the saved reset as recurring capacity. It is a one-off buffer.
- Mixing subscription and API budgets in one line. Subscription seats are capped; API usage is metered. Track them separately so finance can see which one grew.
Bottom line
From 14 September, Claude Code's weekly limits sit 25% above the old standard and 17% below the summer boost. Developers who used more than about 83% of their boosted allowance will now hit the wall. Opus 5.5 added five-hour headroom, a saveable reset, and a more token-efficient top model, but not a bigger weekly envelope. Measure a clean fortnight, split developers into comfortable, tight and blocked, and price the blocked group on both a higher tier and capped API overflow. If you are weighing Claude Code against Cursor, Copilot or Codex at the same time, start with the AI Coding Assistant Finder. The Claude Code harness page has setup details.
Sources (accessed 2026-09-26)
- Hacker News thread quoting @ClaudeDevs (+25% from 14 Sep for Pro, Max, Team, seat-based Enterprise; "17% reduction" compared with the boost): https://news.ycombinator.com/item?id=49491631
- The Economic Times — "Claude Code users to get 17% less weekly usage from September 14 as Anthropic ends 50% boost" (31 Aug 2026; boost through 13 Sep; 29 Aug posts): https://m.economictimes.com/ai/ai-insights/claude-code-users-to-get-17-less-weekly-usage-from-september-14-as-anthropic-ends-50-boost/articleshow/133647006.cms
- paddo.dev — "The Raise That Is a Cut" (4 Sep 2026; boost timeline from 13 May with end dates 13 Jul, 19 Jul, 31 Aug and 13 Sep; no per-plan figures published): https://paddo.dev/blog/the-raise-that-is-a-cut/
- Anthropic — Introducing Claude Opus 5.5 (22 Sep 2026: higher five-hour limits, saveable rate-limit reset, $4/$20, cache reads $0.20, ~40% lower cost than Opus 5; Sonnet 5.5 "in the coming weeks"): https://www.anthropic.com/claude-opus-5-5
- MIXED — Claude Code v2.1.278 server-side auto-mode classifier (21 Sep 2026): https://mixed-news.com/en/claude-code-stops-charging-auto-mode-safety-checks-agents-md/
Editorial note
AI Choice Engine publishes editorial guides to help readers understand fit, trade-offs, and next steps before choosing a tool or provider.
Newsletter
Get the buyer checklist that goes with this guide
Subscribe to download the matching checklist. Automated email delivery is still rolling out — this is not an inbox confirmation.
Updates only — checklists stay free from the resource library, with or without joining. Automated email delivery is still rolling out.