Claude Code Weekly Limits Cut 17% on September 14, 2026
The Short Answer
On September 14, 2026, Claude Code’s weekly limits change in three ways at once:
- The temporary 50% boost ends.
- A permanent 25% increase over the original baseline replaces it, for Pro, Max, Team and seat-based Enterprise plans.
- The separate weekly Opus limit is removed — one pooled allowance, any model.
Against the original baseline, that is a 25% improvement. Against what heavy users have actually had since the boost, it is a reduction of about 17%. Subscription prices do not change.
Last verified: September 3, 2026.
The Arithmetic That Confuses Everyone
The headline number depends entirely on which baseline you measure from, and both numbers are true:
| Baseline | Weekly limit | Change on Sep 14 |
|---|---|---|
| Original standard | 100% | → 125% (+25%) |
| With temporary boost | 150% | → 125% (−17%) |
Anthropic is describing this as a permanent 25% increase, which is accurate relative to the plans as originally sold. Users are describing it as a 17% cut, which is accurate relative to the capacity they have been using. 125 ÷ 150 = 0.833, hence roughly 17% less.
If you subscribed before the boost and never noticed it, September 14 is a straightforward upgrade. If you sized your workflow during the boost period — which is most people who have been running Claude Code heavily this summer — you will feel it as a squeeze.
This Is the Third Limit Change in a Month
The sequence matters, because it tells you what to expect next:
- August 19, 2026 — the earlier weekly limit promotion ended.
- August 31, 2026 — weekly limits were extended again.
- September 14, 2026 — the 50% boost ends, replaced by a permanent +25%.
The pattern is a series of temporary extensions converging on a permanent number. The useful read is that 125% of the original baseline is the number Anthropic intends to keep, and the boosts were capacity management while demand and inference costs settled. Planning against 150% was always planning against a promotion.
The Opus Change Is Genuinely Good
Buried under the limit arithmetic is the structural improvement: the separate Opus weekly limit disappears.
Under the old scheme you had two ceilings — an overall weekly limit and a stricter Opus-specific one. The common failure mode was hitting the Opus ceiling mid-week while still holding plenty of unused general allowance, which forced a downgrade to a weaker model for reasons that had nothing to do with the work.
From September 14 there is one pooled weekly limit, spendable on whatever model you choose. For an Opus-heavy workflow this materially offsets the 17% reduction. For a Sonnet-heavy workflow it is neutral. Nobody is worse off from this specific change.
What to Do Before September 14
Measure your actual burn rate this week. You cannot tell whether a 17% reduction hurts until you know how close to the ceiling you normally run. Most users are not near it; the ones who are, know already.
Stop defaulting to the strongest model. The single largest source of avoidable consumption is routing trivial work — renames, boilerplate, small fixes — to a frontier model out of habit. Reserve the expensive tier for the tasks where it actually changes the outcome.
Use prompt caching properly. Agent loops re-read a large prefix on every turn. Cached reads are billed at a fraction of the input rate, and on Claude Fable 5.1 that multiplier dropped to 0.025x on September 1, 2026. Cache hygiene is the cheapest capacity you will ever buy.
Keep sessions tight. Long-running sessions accumulate context that gets re-processed each turn. Finishing and restarting a session is often cheaper than continuing one.
Have an API fallback. If limits bind, the API has no weekly ceiling — you pay per token. The sensible split is subscription for interactive work, API for batch and background runs. Opus 5 is $5 per million input and $25 per million output tokens; Sonnet 5 is $2/$10 and is permanently priced there after the previously scheduled increase was cancelled.
Should You Switch Tools?
Probably not over this. A 17% capacity reduction at an unchanged price is annoying, not disqualifying, and the pooled-Opus change claws back part of it for the users most likely to complain.
The case for looking around is different: September 2026 has an unusually strong cheap tier. Gemini 3.8 Flash landed on September 2 at $0.75/$3.75 through December, and it is competitive on scoped coding tasks. If a meaningful share of your Claude Code usage is bounded work, moving that share elsewhere relieves the limit pressure without giving up Claude for the work that needs it.
The case against churn: limits move constantly across every vendor, and rebuilding a workflow around a competitor’s promotional pricing is how you end up doing this again in November.