Claude Code's father, the OpenAI network executive, was rejected for a second.

  It's the first time in the history of the world.

"Welcome to anthropic work!"

Just as Claude Code's father, Boris Cherny, was present, in front of hundreds of thousands of people, he extended an invitation to the head of the core product of OpenAI.

It's ridiculous to say the least.

According to the instructions of an OpenAI executive, the hardcore developer stuck the GPT-5.6 Sol into the harness of Claude Code.

The Anthropic opponent is a badge number.

Fifty minutes later, the excavated Tibo refused. Seven minutes later, he did not stop and immediately reset the quota for all paid users.

The all-around Ottman laughed at the X and left a note saying, "lol, one of the things I love about OpenAI most is Tibo".

On July 12, Tibo published an operating manual online.

Five steps, five minutes, a line of alias commands, GPT-5.6 Sol is running in the harness of Claude Code, also known as claudex.

Technically, there is no barrier to launching a local proxy service that hijacks Claude Code's requests and sends them to OpenAI's API.

At the end of the post, he added a provocative sentence, saying that if the plan is sealed, he owes everyone a second time.

I hadn't thought about it, but after six months I was addicted to a word.

On August 8, developer Alex Getman posted a post asking for help.

He moved Tibo's teachings almost in motion, and the code ran shortly afterwards, and Anthropic sealed him directly.

He sent the open source code to GitHub and asked all parties involved for a clear answer, which was not a violation.

Tibo's response was to put anthropic friction on the ground.

I'd love to help you, but I'm not working for Anthropic. It's really weird that they're signing you because you're running another model in your own harness.

The community instantly exploded.

Since then, Claude's sealing system has long been notorious in the developers' circle, with frequent misjudgements, complaints, withdrawals, and paying users to be treated like suspects.

In the comments section, a large number of paying users who share the same illness are popping up.

Some people just paid the subscription fee, the back-foot accounts were closed, the money was locked up, and there was no explanation.

Some complained for 46 days without a voice mail. Others were sealed three months ago, and the work was inundated with stones and no refunds were paid.

What was most disturbing was the extent to which the entire network had to be encompassed, and Anthropic was willing to take a closer look at your complaint.

Scroll down and see

With tears pouring over the commentary, Claude Code's father, Boris Cherny, could not sit down, and the commentary area was extinguished.

He said that the official will never be sealing because users replace the underlying model, and it is almost certain that another account classifier has been triggered incorrectly, is checking, and thank you for making things public.

Soon, Alex Getman's account was restored.

It was less than eight hours.

Boris's phrase "let's go to work with Anthropic" is on the open timeline of X, which is equivalent to Claude Code's head openly digging a corner for his rival generals.

It's a good idea to burn Tibo directly on the fire.

Tibo's response was quite level: "I love the team now, and what I'm going to release in the next few weeks is so exciting".

In a nutshell, not only has OpenAI pledged its loyalty, but it also makes it clear that it still has a huge product base to offer.

Then he chose the most hectic way of celebrating, giving full quotas directly to all paying users.

It's not enough to finish the show, and it's time to put on a show again and again on Monday.

He closed the AI programming circle, relying on "replacement", and hardcore life became a community culture.

To date, he has pressed the reset button 43 times, averaging once every 7.5 days. It has been estimated that Tibo receives a private letter requesting reset every 6 minutes.

Codex surpassed 3 million in April and 10 million in late July, making 1 million a day at its fastest.

To this end, the developers not only built codex-resets.com to track reset records on their own, but also built the TiboGPT website and the Codex Radar tracking app, and even invented the "reset pre-instance" workflow.

It's not a bloodthirsty generosity, it's the codex's steepest weapon of war, a second-rate price battle that airs live on the X-TIME line.

Tibo cries out the same words every time he re-plays: "Come to me, that's enough".

So far, the two companies have demonstrated two completely different user philosophies in these eight hours.

On the one hand, the number limit flows; on the other hand, the limit is full, and there is something to be desired.

Ottman's response, which seems arbitrary, is a strategic concession.

Tibo is one of the best cards OpenAI can play in the AI programming battlefield, relying on its personality charm plus the generosity of real gold and silver.

Another person in charge of someone else's product is the one who wrote the roadmap, and he called the 10 million developers on X his brother and sister.

This technique was not learned by another person, and Ottoman's heart was to understand this and then untie the rope completely.

But the climax of the accident was not unsealed, and the real drama was only played after the accounts were restored.

The developers accidentally hit a barrier while recreating Tibo tutorials.

The core of Claude Code's experience is that it does not come from Claude's model.

MBZUAI's latest study gives accurate figures.

98.4% of Claude Code's functionality comes from harness architecture, and the model itself contributes less than 2%.

From context management on a project to cross-session memory, the core functionality that developers rely on every day is harness.

The model can be pulled down and replaced with a new one at any time, but it carries a harness of memories and habits that can't be removed once worn.

Address of the paper:https://arxiv.org/pdf/2604.14228

Anthropic's tools are modeled after OpenAI, which in turn is likely to be the case.

So this fact overturns the competition logic of AI programming as a whole.

The logic of the past is that the stronger the model, the more users, the more data, the stronger the model gets.

The logic of the Harness era is that the experience of the tool works well, and the user stays here, no matter who runs the model, the provider of the tool is the entrance.

What tools developers use to write code is more important than what models they use.

According to Huang, tokens are moving towards zero profits in commodities, and differentiation is disappearing.

If the model layer really goes down this path, what's really worthwhile is just outside the model.

And this eight-hour drama is the most intuitive show of judging.

Maybe someday in the future people will look back and find that the most intense battlefield in the AI programming circle in the summer of 2026 is not in the model layer.

The winner is the one who has the tools to put the entire workflow of the developer firmly in their own hands.