Finally, Claude Code did something sensible.
Yesterday, the Anthropic developer account @ClaudeDevs announced: When Claude Code reaches its 5-hour usage limit, it will no longer abruptly cut off (mid-code edit, sorry, all progress lost, whatever).
Now, it will first find a proper stopping point, complete the current step, and then stop.
Just this one change blew up the comments section.
Anyone who has used Claude Code knows how nasty this cut is.
It’s not the kind of annoyance where a conversation gets interrupted and you just have to ask again.
You changed a function signature but didn't update the call sites; the tests stopped halfway through, and Claude suddenly halted, leaving you with an incomplete repository—you have to manually check each change to see what has been committed.
Developer @tochukwudev put it most directly: it’s his biggest frustration with Claude Code—every time it gets cut off, it leaves half-finished changes that he has to manually fix. Click Clarify.
To avoid this blow, people came up with all sorts of makeshift solutions.
@PaineofGlass said that when he was close to the limit, he would have Claude save in segments, or simply increase the spending cap. He pays for the peace of mind of not getting interrupted mid-code modification.
Think about it—spending money isn’t about using more of it; it’s about avoiding a painful downfall.
The bell for class dismissal has rung; your teacher has asked you to finish writing this sentence.
This new feature is called Wrap-Up Allowance.
In plain terms, it's the "closing amount."
The mechanism is simple: At the moment you hit the 5-hour limit, the system allocates a small, fixed portion of your weekly quota to allow Claude to complete the current task before stopping.
A message will appear on the screen: “Usage limit reached · wrapping up.”
Like the bell signaling the end of an exam, the teacher asks you to finish writing this sentence before submitting your paper.
That sounds reasonable, doesn't it?
But prior to this, Anthropic’s approach was: the bell rang, and the pen was immediately snatched from your hand—no matter whether you had completed half the exam or had just started writing.
Anthropic now realizes that the previous experience was on the level of a bug.
But don't get too excited—the official help center has drawn clear boundaries:
First, focus only on the current round. Even if you send a new message after crossing the finish line, it will still be blocked—you can only wait for the quota to reset. Want to slip in one more command during the closing window? Sorry, the door is closed.
Second, no additional quota is added. Any remaining usage at the end of the week will still be deducted from your weekly limit. Anthropic has not given you an extra minute of usage—only the way it stops when you reach your limit has been changed.
Third, there is no guarantee that the task will be completed. Once the remaining quota is exhausted, the current step will finish, and the next step will not begin. You can also manually interrupt it at any time.
This quota cannot be applied for, accumulated, or carried over to the next time—it activates automatically when eligible.
If the settlement quota is insufficient, you can activate extra usage to continue. The system will first use up this settlement quota, then charge your usage quota at the standard rate, all while still being subject to your spending limit.
Here's a harsh tiering system.
Pro users: Once per week. Even if you don’t fully use it, that one opportunity counts as used for the week. Over seven days, you get only one chance to stop gracefully—the other six times, you’ll still be cut off abruptly.
Max and Team Premium: Can be used each time the 5-hour limit is reached, as long as the weekly quota has not been exhausted.
Other requirements must also be met: Claude Code must be updated to version 2.1.277 or later, and this applies to the terminal, IDE plugin, or desktop Code interface—but you must be logged in with a Claude account and have an active subscription.
Both the account and the associated organization must have existed for a certain period of time (the exact duration is not disclosed by the platform and may be adjusted in the future). This feature is currently being rolled out in phases and is not available to all accounts.
Neither API key usage nor Amazon Bedrock or Google Cloud Vertex AI can be used. The Claude chat interface is also not applicable.
What is the exact limit for the closing amount? The official statement only mentions that there is a cap depending on the plan, but no specific numbers are provided.
Typical Anthropic style: Letting you know that something exists, but not specifying the exact amount.
Ten days ago, on September 14, the temporary 50% weekly quota promotion for Claude Code expired and was replaced with a permanent 25% increase.
Based on the original amount of 100, it’s a drop from 150 to 125—about 17% lower than the peak.
Anthropic initially only said "permanently increased by 25%" on X, was called out by users, and had a community note added; later, they deleted the post and reposted, finally admitting it was a 17% reduction compared to current levels.
The weekly limit has just been tightened, and the closing limit is also deducted from the weekly limit.
Anthropic hasn't provided any new volume in a minute; the changes only affect how it stops when the quota is exhausted.
Some people switched jobs on the spot, while others poured cold water on the idea.
Comments on X are lively, with clear divisions between supporters and opponents.
@DataDiscovered said that being abruptly cut off mid-edit is the worst failure mode for Claude Code, and this fix represents one of Anthropic's best resolutions to date.
Most directly, @xDqtreats said he’s switching from Codex just because he loves this feature.
@nrajesh4u said he had already reported this request, and likely thousands of others have raised the same issue. In other words, the fact that this feature has only been implemented today is itself a problem.
The ones dampening the enthusiasm are also making valid points.
@shamilkch believes that the next session should pick up from the pre-written checkpoint; otherwise, the fixed closing amount merely makes Claude stop quietly, but it’s still essentially a timeout—just made to look better.
Some are still focused on the 5-hour limit itself. @ysdevastro directly asked: “Can you remove the 5-hour limit for Max?” He said this is the main reason he hasn’t switched over from Codex yet.
Some users are coming to Codex because of this feature, while others are staying on Codex due to the 5-hour limit.
Both companies are competing for the same pool of coders, and the quota rules have long become a key factor developers carefully consider when choosing sides.
Knowing how to stop is also part of the product.
Looking back, this change was really minor.
The 5-hour window has not been extended, the weekly limit has not been increased, and even the small amount of quota used for closing out positions will be deducted from your original weekly limit.
But the comments section is filled with praise. Why?
Because the work AI does has changed.
Previously, people used chatbots for one-off Q&A sessions; if the conversation ended, it ended, and the cost of asking again was nearly zero.
But agents like Claude Code are different—they read multiple files at once, make several code changes, run a round of tests, and iterate again; it’s normal for a single task to take tens of minutes.
Its status isn’t in any single answer—it’s in the middle of this entire sequence of actions. Where you stop determines whether, when you return, you pick up right where you left off or spend half an hour cleaning up the mess.
So Anthropic gave nothing extra, only changed where it stops, and everyone applauded.
Because as AI takes on longer and more complex tasks for people, where it stops and whether you can pick up after it is no longer an issue 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
This article is from the WeChat public account "New Intelligence Yuan," authored by ASI Revelation.