Claude Code weekly limits drop 17% from today's level on Sept 14
The May usage boost expires Sept 13: weekly caps land 25% over the old baseline but 17% under today's. Five-hour windows untouched — here's how to plan.
Copy markdown
The +25% that's also a -17%
Anthropic quotes “+25%,” its critics quote “-17%,” and both are right. Index the old baseline at 100: May's boost lifted weekly Claude Code limits to 150, and Sept 14 resets them permanently to 125 — a quarter above the pre-boost norm, but 125/150 is 17% below what you can burn this week.
What changes for your week
It hits Pro, Max, Team and seat-based Enterprise. Anthropic publishes no absolute token figures, so the only lever is proportional: whatever you max out this week, expect roughly 83% of it after the 14th. Heavy weekly users feel it; light users may never touch the ceiling.
The 5-hour window is safe
Only the weekly ceiling moves — the 5-hour rolling window Anthropic doubled back in May is untouched. Burst-heavy sessions notice nothing; it's the marathon, all-week runs that lose headroom.
Do this before Sunday
Measure your current weekly usage now and budget against 125, not 150. Route cheap work — lint fixes, summaries, boilerplate — to smaller models to reclaim room under the new cap. Anthropic's live limit docs lagged the announcement, so re-check them this week for exact wording.
Elsewhere: Grok 4.7 is aimed at today
xAI has targeted Grok 4.7 — reportedly a ~2.1T-parameter model trained partly on SpaceX data — for Sept 12, but the API docs still list 4.6 as newest, with no model ID or pricing published. Planning a Grok integration? It may be worth waiting a day for real specs.