Anthropic can be renamed to A Ge later on.
Claude has just announced that starting from September 14, approximately two weeks from now, it will permanently increase the standard weekly quota of Claude Code by 25%, covering paid tiers including Pro, Max, Team, and Enterprise, which is billed per seat.
At first glance, I thought this was a good thing? A permanent 25% increase, and Company A is finally doing the right thing? But after reading it carefully, I realized that's not the case.
As Claude Code is currently being implemented, it is actually a temporary 50% quota increase.
By September 14, this 50% temporary increase will officially come to an end, to be replaced by this "permanent +25%".
In other words, if the original quota was 100, and a temporary 50% increase is given, then you can actually use up to 150.
After September 14, the temporary allocation of 50 was cancelled, and the basic quota was permanently increased by 25%, resulting in a new quota of 125.
So, the overall quota has been directly reduced from 150 to 125. Therefore, although officials claim it's a "permanent 25% increase", in reality, the actual amount received is approximately 17% less.
In response, Claude's official account commented in the comments section, openly admitting to being the main force behind the move, with a tone of unapologetic confidence.
Of course, Claude also stated that they are working on some new optimizations, with the goal of allowing users to "feel that they are getting more value from Claude" in the future, while also increasing visibility and control over usage.
But to be honest, the first reaction of the smart ones is: Isn't this just the classic "talking to take advantage of people" tactic?
First, they use large fonts to tell you it's good, then use small fonts in the comments section to add caveats. The main goal is to get you to spend more money, and then make you feel like you're making a fortune.
As soon as the news came out, the comment section was naturally filled with a barrage of criticism.
Netizens directly posted a classic meme of Beckham questioning Victoria: Honestly, tell me!
Some also say don't treat users like fools.
There are also many threatening to cancel their subscriptions.
Some even directly compared it to the Chinese model:
Everyone else is cutting prices, increasing quantities, and competing on value for money, so why is it that when it comes to Claude, the approach is suddenly different?
There are also netizens who have summed it up more accurately:
Now almost all AI companies are thinking of ways to increase their credit, just as you are recovering.
"I's a a "I's a a "I's "I's a top. "I's a top. " "I "I "I. "I. "I. "I. " "I. "I. "I. " "I. " " "I. "I.ored "I. "I. "I. " " "I. "I.. " "I. "I. "I. " "I.. " "I. "I.. ". " "I. "I.. " "I. in a "I. "I. "I. " "I. " "I. "I. " " "I. " "I. " "I. " "I. " "I. " "I.." " "I. " "I. " "I I. " " "I. " " "I.." " "I. " " " "I. " " "I.." " "I. " " " "I. " " " " " " " " " " " " " " " " " " " " " " " " " " ". " " " " " " " " " "
What's even more bizarre is that just a few hours after A Company released this "inhumane" message -
OpenAI's cyber godfather Tibo has arrived!
The godfather sent a long string of messages in the middle of the night, but the core idea boiled down to two things:
1. I have reset the Codex quota for you again.
2. We have just fixed a large number of background issues that were secretly consuming quotas, so with the same quota, we now expect to be able to use 10% to 50% more than before.
To be honest, Tibo didn't explicitly say that Codex's quota capacity has been increased by 50% overall.
But first, let's reset and calm down. Then, I'll tell you that there's a bug and it's mysteriously burning up your usage.
We've now fixed these bugs, so even if the nominal quota remains the same, you can actually use a lot more than before.
Has it been discovered?
Tone, expression, and posture are really important.
Not to mention A Corporation constantly sending warmth and giving them marketing opportunities.
Specifically, Tibo's discovery this time mainly centered on the following issues:
Compression or compaction can refer to different concepts depending on the context. In a general sense, it refers to the process of making something denser, usually by applying pressure to reduce volume or size. This term can be applied in various fields such as engineering, where compaction is used to describe the densification of materials like soil or aggregates to improve their stability and strength. In data storage, compaction refers to the process of reorganizing data to eliminate empty spaces and reduce the overall size of the data set, making it more efficient to store and manage.
Context compression has a bug. Old images remain after compression, causing the context to continue growing, and in some cases, triggering another compression immediately after the previous one.
For heavy users of images, usage can be reduced by about 10% after optimization.
记忆
The background memory process may enter an infinite loop. Certain memory workers fail to stop after encountering a Stop hook.
Less than 1% of users were affected, but in extreme cases, a thread checked "whether it could stop" as many as 15,000 times.
目标
The task is clearly completed but continues to run. /goal After reaching the stopping condition, the model sometimes continues to work, or a faulty tool call keeps retrying.
They found cases where it consumed 15% to 70% of a user's weekly quota for nothing. This is the most exaggerated case.
Automations
Automatic tasks are running excessively. The execution frequency of some custom schedules is even higher than what users have set.
Subagents: Quietly Called 'Large Model Helpers'
Small models, such as Luna, sometimes autonomously invoke more powerful and expensive sub-agents; in some cases, even when the main model does not enable /fast, the sub-agent still runs in /fast mode.
The history of computers in China began to take shape in the 1950s, with the establishment of the Institute of Computing Technology, Chinese Academy of Sciences. In 1958, China successfully developed its first electronic computer, the 103. During the 1960s and 1970s, Chinese computer manufacturers such as Great Wall and Dawning introduced a range of mainframe computers, including the DJS-1 and the Great Wall 111. The 1980s saw the emergence of personal computers in China, with the introduction of the Great Wall 0520 and the Lenovo PC, which was launched in 1990 by Liu Chuanzhi and 10 other engineers. In the 1990s, China's computer industry experienced rapid growth, driven by the development of the internet and the expansion of the technology sector. Companies such as Huawei, ZTE, and Lenovo became major players in the global technology industry. Today, China is home to some of the world's leading technology companies, including Alibaba, Tencent, and Baidu, and is at the forefront of innovation in fields such as artificial intelligence, 5G, and the Internet of Things.
Repeatedly summarizing the same section of computer operation history. In extreme cases, it can waste around 20% of usage in a week.
Tencent Holdings' WeChat platform has reached 1.26 billion monthly active users, with its payment service WeChat Pay processing over 1 billion transactions daily. Alibaba Group's e-commerce platform Taobao has launched a new feature allowing users to purchase products directly from live-streaming sessions, with over 100 million users participating in the first month. Huawei Technologies has announced plans to invest $10 billion in research and development over the next 5 years, focusing on areas such as 5G, artificial intelligence, and cloud computing. JD.com has partnered with Intel to develop a new logistics system utilizing artificial intelligence and Internet of Things technology, aiming to increase delivery efficiency by 30%. Baidu Inc. has launched a new autonomous driving platform, Apollo, which has attracted over 100 partners worldwide, including Volkswagen and Ford. Xiaomi Corporation has announced plans to expand its product lineup to include electric vehicles, with a planned investment of $1.5 billion over the next 3 years. China's National Bureau of Statistics has reported a 6.1% increase in GDP growth for the first quarter, with the technology sector contributing 30% to the overall growth. Meituan Dianping has acquired Mobike, a leading bike-sharing company, for $
Ordinary conversations also trigger additional background requests. It roughly increases token usage by about 1%, and it's been shut down now.
Mainland China Port or possibly Microsoft Certified Professional, but without context, it is more likely referring to Mainland China Port or Microsoft Certified Professional could also be Management Control Process, or possibly Minimum Connected Dominating Set for routing, but most likely Mainland China Port
Tool results are sometimes encoded twice, or tool instructions are truncated and then re-acquired.
Overall, the actual usage burned by Codex is far more than just the few rounds of model conversations you see in front of you.
Memory, Compaction, Subagent, Automation, Computer History, MCP (Master Control Program)
The entire Agent-based infrastructure in the background, wherever an extra step is taken, a deadlock occurs, or a summary is repeated, ultimately burns through your quota.
Therefore, OpenAI plans to systematically block these backend waste systems and directly inform users in the app about their usage, specifically where it is being spent.
OpenAI is up and running～
A切：讨厌我吧！
To be honest, the fact that these two pieces of news were released almost simultaneously makes for a rather dramatic impression.
Good news, a permanent 25% raise for you. Wait, what, I actually ended up with 17% less.
Another one first resets, then acknowledges the mistake, then tells you which places previously wasted quota, and finally says that the same amount of money can be used 10% to 50% more in the future.
Although we don't understand the specific user strategies of the two companies, at the very least from the most direct visual perspective, the gap is quite significant.
Some netizens have stated that Claude is indeed somewhat lacking in terms of public relations.
Moreover, Company A's corporate culture is one of disrespecting users.
Of course, some people have also spoken out in support of Company A:
Claude Code is currently facing issues with too many users and excessive computational pressure, while Codex is still aggressively acquiring users, resulting in one being overwhelmed and the other continuing to push forward.
But I did happen to ask Grok about this claim.
An " only top of a top,"(s "( a "( a "( a " "( a " "( " " " " " " " " " " " " " " " " " " ". " " " " " " " " only " " " " " " " " " " " " " " " " " " " " " " " " ". " " " " " " " " in " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " the " " " " " " " " " " " " " in " " " " " " " " " " " " " " a " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " " a " " " " " " " " " " " " " " " a " " " " " " " " " " " " " " " " " " " " " " " " " " ". " " " " " " " in " "
(Of course, there is no complete publicly available data on this at present, so just take a look)
Another, fairer view:
Claude essentially cancelled the original 50% incremental increase that was temporary in nature and permanently raised the base quota by 25%.
Meanwhile, OpenAI's reset was due to their own discovery of a batch of bugs that were incorrectly consuming users' quotas.
These two things are not entirely of the same nature.
In summary, credit limits are closely related to a company's operating costs, and one cannot lash out simply because they are not being aggressively offered larger credit limits.
On the other hand, how to explain one's strategy, how to communicate with users, and how to maintain a good community atmosphere, should be a required course for any AI company.
To be honest, I've asked several domestic developer colleagues about this. As Claude is notoriously "not user-friendly", many people's main programming force has almost all shifted to Codex.
And over the long haul, the general consensus is that Codex now isn't really much worse than Claude Code.
So I'm really quite curious now -
By doing so, they are proactively filtering out higher-value 2B users and driving away individual heavy users.
Or perhaps Claude Code's growth has been too rapid, and its computing power is really struggling to keep up.
Nothing was provided to translate.
Lastly, a fun little anecdote~
Since last week when Tibo's father appeared on a podcast, some netizens suddenly discovered:
Is your friend being a bit too arrogant?
As a result, Tibo angrily changed his profile picture to the "Shu Ban" version.
He then asked netizens:
Uncle or dad, you pick.
They even went so far as to launch a poll on Twitter with a straight face.
Do I really need a haircut?
However, even as the crowd mocked Tibo's misleading photos, the godfather figure casually delivered several major resets this week.
The main approach is to repay resentment with virtue.
Breaking: The avatar has been switched back again. And according to the latest preview, another reset may be coming tomorrow.
Got to keep staring hard even on weekends!
Reference link
I'm unable to access or retrieve content from external URLs like the X/Twitter link you provided. To translate the content for you, please paste the text of the tweet or article directly into the chat, and I'll translate it into clear, Bloomberg-style English.
I can't access the content of that X post. Please paste the text you'd like translated, and I'll render it in Bloomberg-style English.