Claude Code reduces it's weekly limit by 17% – compared to today | AI Paper Digest

Claude Code weekly limits rise 25% permanently on Sept 14, but that is a 17% cut from today's.

Developers on Claude Pro or Max who use Claude Code daily and worry about weekly limit changes

Permanent 25% limit increase from Sept 14
- It covers Pro, Max, Team, and seat-based Enterprise plans.
- Standard weekly limits in Claude Code go up 25%.
50% temporary boost stays until then
- The current 50% increase remains in place until September 14.
Net 17% cut versus today
- ClaudeDevs said weekly limits work out to a 17% reduction compared to today.
- The limit moves from a temporary 50% boost to a permanent 25% boost.
Better visibility and control promised
- They are working on changes to make it feel like you get more from Claude.
- They did not share specifics.
Still finding a sustainable level
- ClaudeDevs said they were figuring out what they can sustainably serve going forward.

Felt closer to 30% than 17%
- A $100 subscriber said they started getting out-of-credit notices two weeks ago.
- They expect to hit limits constantly once the boost ends.
Permanent 25% beats temporary 50%
- One commenter compared it to a permanent raise.
- They said they never reached even 50% of the boosted limits.
Mixing other models cut costs
- One user said Claude Pro limits hit constantly with Claude Code.
- They now combine Kimi, DeepSeek and others on a $10 OpenCode Go plan with hundreds of millions of mostly cached tokens daily.
- They credited good AGENTS.md and a short ARCHITECTURE.md.
Complaints about communication and churn
- Some criticized not saying 17% cut plainly.
- Others complained a $200 plan changes every two weeks, and one said they could switch to Z.ai/GLM.

Check last week's Claude Code usage
- Record the weekly limit percentage you used today.
- Calculate whether you fit within 83% of the current limit.
Tidy AGENTS.md and ARCHITECTURE.md
- Summarize project structure in a short ARCHITECTURE.md.
- Reduce wasted exploration by the agent.
Try a small task on an alternative model
- Test one other model on the same codebase in case you hit limits.
- Compare results and cost on the same task.

Weekly limitA cap on usage per week, like a monthly data plan on a weekly cycle.

Promotional limitA temporarily raised limit that returns to normal when the period ends.

AGENTS.mdA file telling coding agents the project rules, like an onboarding doc for a new hire.

Cache hitA token served from previously processed input, like not rereading a book you already read.

Claude Code reduces it's weekly limit by 17% – compared to today