Claude Code's Permanent +25% Is a 17% Cut

“Permanently raised by 25%” and “a 17% reduction” describe the same event.

When Anthropic announced the change to Claude Code’s weekly limits at the end of August, both numbers appeared at once and the community argued for days. The version I wrote then went: Anthropic first dressed it up as an increase, users worked out that it was a cut, and only then did it post an admission.

This time I captured that thread and looked, and found that account isn’t accurate. The two posts went out in the same minute. Anthropic wrote the 17% down from the start.

That detail changes how the whole thing should be read, so I’ve rewritten it.

What did Anthropic actually say?

Here’s the thread:

Figure: the ClaudeDevs thread of August 29 (US time), both posts with the same timestamp (source: x.com)

The top post is the conclusion: from September 14, standard weekly limits in Claude Code are permanently raised by 25%, for Pro, Max, Team and seat-based Enterprise plans. Until then, the current +50% stays in place.

The one below is the addendum: compared to today, this works out to a 17% reduction in weekly limits. It then previews changes in the works that will make it feel like you’re getting more from Claude, while having more visibility and control of your usage.

The timestamps on the two posts in the figure are the same minute. This is a thread written in advance and sent together; the second post wasn’t added afterwards.

So I need to correct what I wrote in August. The original said Anthropic “didn’t dig in when challenged and followed up shortly after”. In fact there was no “shortly after”, and it can’t really be called forced out of them.

I don’t want to overstate it the other way either. The headline-level post is the one about +25%, and that’s the one that gets screenshotted, shared and turned into headlines. The “−17%” sits in the reply. The information is complete; where it was placed was chosen. BleepingComputer’s headline at the time said simply “Anthropic is cutting Claude Code’s current weekly limits by 17%”, which amounts to moving the second post into the first post’s position on readers’ behalf.

How can both numbers be right at once?

Because they’re compared against different things.

Take the weekly limit before the promotion as 100. The promotion that began May 13 added 50%, making it 150. After September 14 it’s 125.

Figure: the conversion with the original limit set to 100 (calculated for this article from the percentages ClaudeDevs published)

125 is 25% more than 100. That’s “permanently raised by 25%”.

125 is 25 less than 150, and 25 divided by 150 is 16.7%, which rounds to 17%. That’s “a 17% reduction”.

There’s a small arithmetic trap here that many people get wrong the first time. Add 50% and then take off 25%, and intuition says the net is +25%, so falling back from the promotion level should be a 25% drop. It isn’t. The base of the percentage changed. Going from 150 to 125 loses a sixth of 150.

One general rule keeps you from getting it wrong again: after adding x, the proportion you have to remove to get back to the start is x divided by (1 + x). After adding 50% you’d need to remove 33% to return to the start, and after adding 100% you’d need to remove 50%. The next time you see any add-then-subtract announcement, run that formula to see what the real difference is.

Which number is “more real”? It depends on where your memory starts.

If you only began using Claude Code after this May, you’ve never seen 100, and your normal is 150. For you this is a reduction, and there’s no other way to read it.

If you’ve been using it since earlier, you remember what 100 felt like, and 125 really is more than what you originally paid for.

Anthropic chose the baseline that flatters it for the lead sentence, which is normal; every company does that. What I care about more is something else: that baseline is also defined by Anthropic. The promotion was extended four times and lasted four months, a sequence I’ve laid out separately. When something is provided for four months, it’s entirely reasonable for users to treat it as the norm. Saying in the fourth month “that was only a promotion” holds up in the terms and is hard to hold up in how it feels.

What does the official documentation say now?

The August article had a passage saying the support center’s article hadn’t been updated and the latest policy could only be seen in posts. It has been updated now. I looked at the current version, and a note has been added at the top:

Figure: the current opening note of the support article (source: support.claude.com)

This passage contains a piece of information that wasn’t in the posts: “We were planning to return weekly limits to their original levels.”

In other words, under the original plan it should have been 100 after September 14, not 125. That 25% was a later decision to keep.

That gives the whole thing one more reading. From Anthropic’s side, they were going to take all of it back and in the end left you half. From the user’s side, they gave you 150 for four months and then took away a sixth. Both describe the same set of numbers.

I noted one other sentence from the article: the full promotional boost “couldn’t last”. It echoes the mid-August post about capacity possibly being tight.

Who notices, and who isn’t affected at all?

This change only affects some people. What matters is how much you actually used during the promotion.

Figure: the effect after September 14 by actual use during the promotion (derived for this article with the original limit set to 100)

If your weekly usage never went above 125 in those four months, you won’t feel anything after September 14. And compared with the period before May, your ceiling is 25% higher.

If you often went above 125, you’ll hit the wall. The fuller you ran, the bigger the difference. For people who regularly used up the full 150, what’s lost each week is a sixth of their work.

The trouble is that most people don’t know which row they’re in. The support center mentions that typing /usage in the CLI shows current usage and limits. If you didn’t note your numbers during the promotion, all you can do now is infer from whether you’ve been hitting the cap more often since September 14.

Two more things to remember, both from that support article. This change touched only the weekly limit; the 5-hour limit never changed at any point. And it covers only Claude Code; regular Claude chat allowances are outside its scope. So if the layer that usually stops you is the 5-hour one, none of this concerns you.

Has the promised “visibility and control” shipped?

Both the post and the support article preview a follow-up: more features around usage, visibility and control.

As I write this in mid-October, I searched around and found no announcement that these features have formally launched. Some outlets describe them as new dashboard tools, but that’s a paraphrase of the preview, not something already released.

The only related move I could find is the usage reset coupon with an expiry date that came when Claude Opus 5.5 launched on September 23. That looks more like a one-off compensation than the kind of lasting feature the preview describes.

I’ll keep watching this. “Make it feel like you’re getting more” is neatly worded; it’s about a feeling. If what arrives is a clearer usage display, that’s really useful, but it won’t turn 125 back into 150.

How does this differ from what OpenAI did?

The timing makes for a comparison. In the same month, OpenAI took the $200 ChatGPT Pro plan off sale for 20 days, and when it went back on sale the price was unchanged while its own product lead said it nets out at about half the old plan in API spend.

The two companies did the same kind of thing: price unchanged, allowance lowered, with compute as the reason both times. The difference is in the size and the manner.

Anthropic cut 17%, said so two weeks ahead, and put the 17% in writing. OpenAI cut about half, first shutting the entrance abruptly and then coming back with a new calculation. Both gave existing users a buffer: Anthropic kept the +50% through September 13, and OpenAI kept the old allowance through October 29.

My view is that on this occasion Anthropic handled it somewhat better, and the reason isn’t that it cut less but that it wrote the ugly number out itself. Plenty of companies would have posted only the first post.

What would make me take that back? If the previewed features keep not appearing, or if the next adjustment no longer comes with a number compared against “today”. Writing bad news out clearly only counts if it’s kept up.

What can you do now?

1. Run /usage once and note what fraction of the weekly limit you currently use. That’s your baseline for judging any future adjustment.
2. Work out which layer stops you. If it’s the 5-hour limit, no change to the weekly limit matters to you, and what you should look at is spreading your work out.
3. Don’t budget on promotions. Treat an increase with an end date as if it didn’t exist. The lesson here is that even a promotion extended four times still ends.
4. If you really need 150, the only options are upgrading your plan or moving to the pay-as-you-go API. Both cost more, but at least the allowance is one you bought.

If the billing terms are unfamiliar, the usage-limits entry in the glossary covers them.

What this rewrite corrected

• The original said Anthropic admitted the reduction by “following up shortly after” being challenged. The two posts are one thread, sent in the same minute.
• The original said the support center’s article hadn’t been updated. It has been, and gained the sentence about having planned to return to original levels.
• The original cited the mood of the discussion on Hacker News. I didn’t re-verify that this time and removed it.
• The 100/150/125 in the tables is a conversion for illustration, not actual allowance figures published by Anthropic. It published only percentages.
• I didn’t measure the allowance before and after September 14 myself; this article relies entirely on the official posts and the support article.