Finally, Claude Code did something decent.
Just yesterday, Anthropic's developer account @ClaudeDevs officially announced: when Claude Code hits the 5-hour usage limit, it no longer cuts off abruptly (code half-changed, sorry, it's gone, deal with it however you like).
Now, it will first find a decent stopping point, finish the current step, and then stop.
Just this one change, and the comments section exploded on the spot.
Anyone who has used Claude Code knows how disgusting this cut-off is.
It's not the kind of disgusting where a chat gets interrupted and you just ask again.
It's that you changed a function signature, the call sites haven't been updated yet, the tests are halfway through running, and Claude suddenly stops. It leaves you with a half-finished repository, and you have to go through it spot by spot to figure out which changes have already been written to disk.
Developer @tochukwudev put it most directly: this is his biggest annoyance with Claude Code, every time it gets cut off it leaves half-finished changes, and he has to manually sort through them one by one
Diandian
.
To dodge this cut-off, people have come up with all kinds of crude workarounds.
@PaineofGlass said that when he used to get close to the limit, he would have Claude save its progress in several segments, or simply raise the overage spending cap. Paying money to buy a "don't cut off in the middle of changing code."
Savor that: you're not paying to use more, you're paying so it doesn't die too ugly.
The bell rings, and the teacher lets you finish writing this sentence
This new feature is called Wrap-Up Allowance.
In plain terms, it's a "wrap-up allowance."
The mechanism is simple: the moment you hit the 5-hour limit, the system carves out a small fixed allowance from your weekly quota, letting Claude finish the step it's currently doing before stopping.
A line of text will appear in the interface — "Usage limit reached · wrapping up."
It's like when the exam bell rings and the teacher lets you finish writing the sentence you're on before handing in your paper.
Sounds very reasonable, right?
But before this, Anthropic's approach was: the bell rings, the pen is yanked straight out of your hand, whatever you've written is what you get, no matter whether you're halfway through the paper or just started.
Anthropic finally realizes now that the previous experience was bug-level.
But don't celebrate too early, the official help center draws the boundaries very clearly:
First, it only covers the current running turn. After you hit the limit, if you send a new message, it will still be blocked, and you can only wait for the quota to reset. Want to sneak in one more instruction during the wrap-up window? Sorry, the door is closed.
Second, no extra amount is added. The quota spent on wrapping up is still deducted from your weekly limit as usual. Anthropic doesn't give you one extra minute of usage; what it changed is only how it stops at the moment of running out.
Third, it doesn't guarantee finishing the job. Once the wrap-up allowance is used up and the current step is done, the next step will no longer begin. You can also manually interrupt it at any time.
This allowance cannot be applied for, cannot be saved up, and cannot be carried over to the next time. It takes effect automatically once you meet the conditions.
If the wrap-up isn't enough, you can enable extra usage and keep going. The system first uses up this wrap-up allowance, then deducts from your usage allowance at the standard price, and it's still governed by the spending cap.
Here's a brutal tier breakdown.
Pro users: at most once per week. Even if you don't use it up that one time, it still counts as used for the week. In other words: out of seven days a week, you only get one chance to "stop with dignity." The other six times you hit the wall, you still get cut off just the same.
Max and Team Premium: you can use it every time you hit the 5-hour limit, provided your weekly quota hasn't been exhausted.
There are also a few other thresholds to meet: Claude Code must be upgraded to version 2.1.277 or newer, and it works in the terminal, IDE plugins, and the Code page on desktop, but you must be logged in with a Claude account subscription.
The account and the organization it belongs to each need to have existed for a certain period of time (the official length is not disclosed and may be adjusted later). The feature is currently being rolled out in batches, so not every account has it.
Calls made with an API Key, or through Amazon Bedrock or Google Cloud Vertex AI, cannot use it. The Claude chat window also doesn't apply.
How much wrap-up allowance is there exactly? The official word is only "it varies by plan and has a cap," with no numbers given.
Typical Anthropic style: they tell you this thing exists, but don't ask exactly how much.
Rewind ten days. On September 14, the temporary 50% weekly quota promotion that Claude Code had added earlier expired, replaced by a permanent 25% increase.
Taking the original quota as 100, that's going from 150 back to 125—about 17% less than at the peak.
At first, Anthropic only said on X that it was a "permanent 25% increase." After users pushed back and a community note was attached, they deleted the post and reposted, only then honestly admitting that "compared to now, this is a 17% reduction."
The weekly quota had just been tightened, and the wrap-up allowance is deducted from the weekly quota.
Anthropic didn't add a single bit of new quota; what it changed is purely how things stop the moment the quota runs out.
Some people "jumped ship" on the spot, others poured cold water on the spot
The comments on X were lively, and the pro and con camps were clearly divided.
@DataDiscovered said that being hard-cut mid-edit is the worst failure mode of Claude Code, and this wrap-up allowance counts as one of Anthropic's best fixes to date.
The most direct was @xDqtreats, who said he is switching over from Codex just because he likes this feature.
@nrajesh4u said he had given feedback on this need long ago, and estimated that thousands of others had raised the same thing. In other words, the fact that this feature only arrived today is itself a problem.
Those pouring cold water also have a point.
@shamilkch believes that the wrap-up only makes sense if the next session can continue from a written checkpoint. Otherwise, this fixed wrap-up allowance just lets Claude stop more quietly; in essence it's still a timeout, just a prettier one.
Others are fixated on the 5-hour limit itself. @ysdevastro directly called out: can you remove the 5-hour limit for Max? He said this is the main reason he hasn't switched over from Codex.
On one side, some people are coming over from Codex because of this feature; on the other side, some people are staying with Codex because of the 5-hour limit.
Both are competing for the same group of people who write code, and quota rules have long become something developers carefully compare when choosing sides.
How it stops is also part of the product
Looking back, this change was really small.
The 5-hour window was not widened, the weekly quota was not increased, and even the bit of quota used for wrap-up still has to be deducted from your original weekly limit.
But the comment section was full of praise. Why?
Because the work AI does has changed.
In the past, people used chatbots, one question and one answer, and if it was cut off, it was cut off; the cost of asking again was close to zero.
But an Agent like Claude Code is different - it reads several files at once, changes several places in the code, runs a round of tests and then changes again, and it is normal for one task to run for several tens of minutes.
Its state is not in some single answer, but in the middle of this entire string of actions. Where it stops determines whether, when you come back to take over, you continue working or first spend half an hour cleaning up the mess.
So Anthropic did not give anything extra, and only changed where it stops, and everyone praised it.
Because as the work AI does on people's behalf becomes longer and more complex, where it stops and whether you can continue working after it stops is no longer a question of "experience optimization" -
it is the product itself.
References:
https://x.com/ClaudeDevs/status/2103561342057943314
https://support.claude.com/en/articles/17040437-claude-code-wrap-up-allowance
This article comes from the WeChat public account "Xin Zhi Yuan", author: ASI Revelation
