Abstract:
Finally, Claude Code did something. Just yesterday, Anthropic developer account @ClaudeDevs officially announced: Claude Code will no longer be cut off when it hits the 5-hour usage limit (the code is changed halfway, sorry, it’s gone, what the hell).
Now, it will first find a decent stopping point, complete the step on hand, and then stop.
With just such a change, the comment section exploded on the spot.
Anyone who has used Claude Code knows how disgusting this knife is.
It’s not the kind of disgusting situation where the chat is interrupted and you just have to ask another question.
You changed the signature of a function, but the calling place has not been changed. Halfway through the test, Claude suddenly stopped. You are left with a half-warehouse, and you have to go through it yourself to check which changes have already been made.
Developer @tochukwudev put it most directly: This is his biggest worry when using Claude Code. Every time it is cut off, half of the changes are left, and he has to manually sort it out bit by bit.
In order to avoid this knife, everyone practiced various methods.
@PaineofGlass said that he used to ask Claude to save in several segments when he was close to the limit, or simply raise the limit for excess spending. Spend money to buy a "don't break in the middle of changing the code".
You Pinpin, you spend money not to use it more, but to avoid an ugly death.
The closing bell rang and the teacher asked you to finish writing this sentence
This new feature is called Wrap-Up Allowance.
Translated into adult language, it is the "ending quota".
The mechanism is very simple: the moment you hit the 5-hour limit, the system will draw a small fixed amount from your weekly quota and let Claude finish the step he is doing before stopping.
A line of prompts will appear on the interface - "Usage limit reached · wrapping up."
It's like the exam bell rings, and the teacher asks you to finish the sentence you are writing before handing it in.
Sounds reasonable, right?
But before that, Anthropic's approach is: when the bell rings, the pen is taken directly from your hand, and you finish writing wherever you go, whether you are halfway through the paper or just started writing.
Anthropic now finally knows that the previous one was a bug-level experience.
But don’t be too happy too early. The official help center draws the boundaries very clearly:
First, just run this round.
After hitting the line, if you send a new message, it will still be blocked, and you can only wait for the limit to be reset. Want to take advantage of the closing window to squeeze in an extra instruction? Sorry, the door is closed.Second, no additional dosage is required.
The final amount spent will be deducted from the weekly limit as usual. Anthropic doesn't give you an extra minute, it just changes how to stop when you use up.Third, there is no guarantee that the work will be completed.
When the closing quota is used up and the current step is completed, the next step will not start again. You can also interrupt it manually at any time.
This quota cannot be applied for, saved, or moved to the next time. It will automatically take effect if the conditions are met.
If finishing is not enough, you can turn on extra usage and continue running. The system will first use up the closing quota, and then deduct your usage quota according to the standard price, which is still subject to the consumption limit.
Here is a heart-wrenching sub-category.
Pro users:
Maximum of once a week.
Even if it wasn't fully used that time, it could still be considered used this week. Translation: Seven days a week, you only have one chance to "stop gracefully", and the remaining six times you hit the wall, you will still be cut off.Max and Team Premium:
You can use it every time you hit the 5-hour limit, provided that the weekly quota has not been exhausted.
Several other thresholds must also be met: Claude Code must be upgraded to version 2.1.277 or newer. The terminal, IDE plug-in, and desktop Code page are all acceptable, but you must log in and subscribe with a Claude account.
The account and the organization must exist for a full period of time (this period is not officially disclosed and may be adjusted in the future). The function is still being released in batches, and not every account has it.
It won’t work if you use API Key or use Amazon Bedrock or Google Cloud Vertex AI. The Claude chat window doesn't work either.
How much is the closing amount? The official only said that "it varies according to the package and has a cap," without giving a figure.
Typical Anthropic style: tell you there is this thing, but don’t ask how much.
Going forward ten days, on September 14th, Claude Code’s temporary 50% weekly quota promotion expired and was replaced by a permanent increase of 25%.
Based on the original amount of 100, it means returning from 150 to 125 - about 17% less than the maximum amount.
Anthropic initially only said "a permanent increase of 25%" on
The weekly quota has just been tightened, and the closing quota is deducted from the weekly quota.
Anthropic didn’t give out new quotas for even one minute. The only change was how to stop when the quota is used up.
Some people "jumped" on the spot, while others poured cold water on the spot
The comments on X are very lively, and the pros and cons are clearly divided.
@DataDiscovered said that being hard-cut mid-editing is the worst way for Claude Code to malfunction, and this closing amount can be regarded as one of Anthropic's best fixes so far.
The most direct one is @xDqtreats. He said that he is switching from Codex because he likes this feature.
@nrajesh4u said that he had already reported this requirement, and it is estimated that thousands of people have mentioned the same thing. In other words, the fact that this feature was not released until today is a problem in itself.
Those who throw cold water on it are right.
@shamilkch believes that the next session can continue from the written checkpoint, and the ending will be meaningful. Otherwise, this fixed closing amount will just allow Claude to park more quietly. It is essentially a timeout, but it looks better.
Some people also focus on the 5-hour limit itself. @ysdevastro shouted directly: Can the 5-hour limit be lifted for Max? He said this was the main reason why he hadn't switched from Codex yet.
Some people come from Codex because of this feature, and some people stay in Codex because of the 5-hour limit.
The two companies are competing for the same group of people who write code, and the quota rules have long become a serious comparison for developers when choosing sides.
How to stop is also part of the product
Looking back, this change is really small.
The 5-hour window has not been relaxed, the weekly quota has not been increased, and even the quota used at the end will be deducted from your original weekly quota.
But the comment section was full of applause. Why?
Because the work done by AI has changed.
In the past, people used chatbots to ask and answer questions, and then they were interrupted. The cost of asking again was close to zero.
But an Agent like Claude Code is different - it reads several files at a time, changes several codes, runs a round of tests and then makes changes. It is normal to run a task for dozens of minutes.
Its status is not in a certain answer, but in the middle of the entire series of actions. Where you stop determines whether you continue to work when you come back to take over, or whether you spend half an hour cleaning up the mess.
So Anthropic didn’t give anything more, it just changed the stopping position, and everyone applauded.
Because when
AI
The work done for others is getting longer and more complex. Where it stops and whether you can continue to do it after it stops is no longer a question of "experience optimization"——
It is the product itself.
Reference materials:
https://x.com/ClaudeDevs/status/2103561342057943314
https://support.claude.com/en/articles/17040437-claude-code-wrap-up-allowance
Comments