WHAT TO DO TODAY, IN ONE LINE
● Run /usage and write down what it says. That number disappears at midnight PT.
● The boost ends at 11:59pm PT tonight, after being extended four times.
● Tomorrow: roughly 17 percent less than today, if the reported figure holds.
● Why it matters: you cannot measure a loss after the thing is gone.
Why today rather than tomorrow
Most of the coverage of this change has been arithmetic — 100 before May, 150 today, 125 tomorrow. That is accurate and it does not tell you whether you personally are affected.
The only thing that tells you is your own consumption against today's ceiling. And today is the last day that ceiling exists.
THE NUMBER THAT MATTERS IS 83 PERCENT
Today sits at 150 on the index. Tomorrow is reported at 125. So anything above 83 percent of today's weekly bar exceeds tomorrow's ceiling.
Below that, this change does not affect you at all and you can stop reading. Above it, you will hit the new limit and it is worth knowing now rather than mid-task next Thursday.
Three things to record
| What |
Why |
| Your current weekly usage percentage | The baseline for everything below |
| Your worst week in the last month | Average weeks are not what breaks. Peak weeks are |
| What you were doing in that week | If it was one large refactor, that is a different problem from steady use |
The thing most people will miss
On Pro and Max, Claude Code shares its weekly allowance with chat and Cowork. So your usage figure is not a coding figure — it is everything you did with Claude that week.
Which means a heavy Cowork week and a heavy coding week compete with each other, and after tomorrow they compete inside a smaller pool. Worth knowing which of the three is actually consuming your allowance before assuming it is the coding.
What is not changing
- Five-hour limits. Doubled separately in May, unaffected by any of this.
- API and usage-based billing. The promotion never applied.
- The free plan. Never eligible.
- Model access or capability. Claude Code is the same product tomorrow as today.
If you are above 83 percent
- Trim before you upgrade. Stop sending full context every step, cap diff sizes, restart long sessions from a handover. That usually recovers more than 17 percent and costs nothing.
- Find out which product is eating the pool. Chat, Code and Cowork draw from the same allowance.
- Price a month of API before paying for a higher tier. At Sonnet 5 rates, $100 buys roughly 33 million input tokens.
- Set a per-run spend cap if anything runs unattended. A weekly pool is the wrong shape for a loop.
One note on the number
The 17 percent figure comes from Anthropic's own clarification after it deleted and reposted the original announcement. The replacement level of 25 percent above baseline appears in that ClaudeDevs post and, as of today, on no Anthropic-owned documentation page — the Help Center states only that limits return to their standard levels.
That is another reason to measure rather than calculate. Run /usage tomorrow too, and compare against what you wrote down today.
Sources
FAQ
When exactly does the boost end?
11:59pm PT on 13 September 2026. The promotion ran from 13 May and was extended four times from its original 13 July end date.
How do I check my usage?
Run /usage in Claude Code. Write the number down today — it is the only day you can see what the higher ceiling looked like.
What is the 83 percent figure?
Today sits at 150 on the index and tomorrow is reported at 125. Anything above 83 percent of today's weekly bar would exceed tomorrow's ceiling.
Does this affect my five-hour limits?
No. Those were doubled separately in May and are unaffected.
Does it affect API usage?
No. The promotion applied to subscription plans only.
Is the replacement figure confirmed?
The 25 percent appears in the official ClaudeDevs post on X but on no Anthropic documentation page, which states only that limits return to standard levels. Measure rather than calculate.