OpenAI, which is increasingly similar to the Taiping Heavenly Kingdom, has rolled out GPT-6
This morning, OpenAI launched GPT-6 Astra, the most powerful model in its history, declaring "Welcome to the AGI Era".
The company that wrote "ensuring AGI benefits all of humanity" into its charter has finally ushered in its own "kingdom of heaven".
Yet, it itself is increasingly resembling the Taiping Heavenly Kingdom.
The Chaos Before the Arrival of AGI
Everything seems far too coincidental.
Last night, ChatGPT, Claude, and Grok all experienced simultaneous outages for a period of time. It felt as if something major was about to happen.
Soon, OpenAI released GPT-6 Astra. It is described as "the most capable model it has ever deployed at scale", with the ability to directly operate computers, browse web pages, develop software, conduct scientific research, and handle complex professional tasks.
At the media conference, OpenAI President Brockman said: Welcome to the AGI Era.
For a company that has embedded AGI in its mission since the very first day of its founding, this statement carries extraordinary weight.
Eight days before the release of Astra, *Time* magazine just published a feature story on OpenAI. The author stayed at OpenAI's headquarters for two consecutive weeks in August, interviewing more than 20 company executives, employees, clients, investors, and competitors.
The report did not portray a triumphant enterprise advancing triumphantly all the way, but a company that has just gone through a year of chaos and is now attempting to "restart".
Executives have left in succession, the enterprise market is being chased by Anthropic, some product lines have been scaled back, and Altman himself has admitted that the company made mistakes in products, pre-training and research over the past year.
In the same interview, OpenAI's senior management was more radical about the future than ever before. Altman stated that by the end of 2026, the company may have an internal system that he himself would be willing to call AGI.
Altman's "end of the year" has not yet arrived, but the slogan of the "AGI Era" has come first. OpenAI seems to be a little too eager, after all, the turmoil the company has gone through is obvious to all.
According to public reports, since the beginning of 2026, at least 14 senior managers, business leaders or core team leads at OpenAI have left the company or stepped down from their original full-time positions.
In just the past two months, OpenAI has lost its second-in-command, Chief Revenue Officer, Chief Futurist, Head of Safety Systems, senior safety researcher, and its only full-time ethicist all at the same time.
Brockman explained on camera at CNBC: "The company has gone through different stages of development, and each period has a different management team. Sam and I have always stayed at the company, and what truly makes us strong is precisely the complexity and diversity of the people on our team."
It is a decent response. However, there were 11 people on OpenAI's founding team, and now only he and Altman remain among the founding members who hold the core daily operational rights of the commercial company.
This wave of departures also took place after the IPO window had been opened. On June 8, 2026, OpenAI confirmed that it had confidentially submitted its S-1 form to the U.S. Securities and Exchange Commission. The company did not commit to a specific IPO timeline.
Logically, this should be the easiest time to retain talents. But the golden handcuffs have long been far less solid than they appear.
OpenAI has previously provided employees with multiple cash-out windows, and some people preparing to leave seem to have cashed out part of their wealth in advance through company repurchases and secondary transactions. After leaving OpenAI, they may become the founders of a new lab, with larger equity stakes and a steeper personal growth curve.
OpenAI is clearly no longer their ideal country.
OpenAI and the Taiping Heavenly Kingdom
OpenAI is increasingly resembling the Taiping Heavenly Kingdom. At the very beginning, both of them gathered talents with missions that transcended mundane interests.
The Taiping Heavenly Kingdom promised "one community under heaven", where all land is cultivated together, all food is shared, and no one goes without food and warmth; OpenAI wrote in its charter that it would "ensure AGI benefits all of humanity", and its primary responsibility is not to shareholders, but to humanity as a whole.
The former's "peace and prosperity" cannot be quantified, and the latter's "benefiting all of humanity" cannot be quantified either. Precisely because their goals are grand enough, they possess the mobilization power that ordinary organizations do not have: people are willing to set aside their immediate real interests temporarily, believing that they are participating in something that will change the world.
The Taiping Heavenly Kingdom also put forward two promises: the sacred treasury system where "no one holds private property", and the *Taiping Heavenly Kingdom Land System* that ensures "no one is left without food and warmth". OpenAI also has two such commitments: the non-profit structure that "is not constrained by financial returns", and the charter that states "the primary responsibility is to all of humanity". Both were announced with great fanfare, but faded into silence when it came to implementation.
After the Jintian Uprising, the Taiping Heavenly Kingdom established its administrative system and enfeoffed various princes in Yong'an. The Hong King was the nominal leader, but the real power was held by the East King.
At OpenAI, Altman is the Hong King, in charge of fundraising, commercialization, and public narratives, with power derived from "I am leading this company to fulfill its mission". The East King corresponds to Ilya Sutskever, the Chief Scientist and widely recognized technical authority in the AGI field, who was the closest to the figure that could interpret the "will of the Heavenly Father".
In 1856, the Tianjing Incident broke out. Yang Xiuqing forced Hong Xiuquan to confer the title of "Wansui (the Immortal)" on him, Hong Xiuquan secretly summoned Wei Changhui back to the capital, and Wei Changhui massacred Yang Xiuqing's entire residence. Shi Dakai was suspected and left the capital with his troops, marking the point where the Taiping Heavenly Kingdom went from prosperity to decline.
OpenAI also had its own Tianjing Incident: in November 2023, the OpenAI board of directors suddenly removed Altman from his position, and Ilya was one of the leaders of this coup.
The coup only lasted five days, more than 700 employees signed a joint letter demanding the board of directors step down. Then Ilya defected, stating that he "deeply regretted his participation in the board's actions". In the end, Altman made his "triumphant return", and the coup faction was purged. Ilya left the company, the entire safety team was ousted, and Altman held absolute sole power.
The internal strife of the Taiping Heavenly Kingdom did not only happen once with the Tianjing Incident, and OpenAI's purges did not happen only once either.
The earliest one involved Elon Musk, a co-founder, who left in 2018 and later became hostile to the company. In 2021, due to disagreements over the safety roadmap, Dario Amodei and his sister left and founded Anthropic. In 2024, the Super Alignment team was disbanded, and a group of safety researchers scattered to other organizations; in 2026, the wave of executive departures took place.
Three years after the Tianjing Incident, Hong Rengan arrived in Tianjing from Hong Kong and put forward the *New Outline for Government Administration*, a complete modernization plan, but unfortunately it was too late, and most of it was never implemented. The commercialization system represented by people like Fidji Simo was truly established, but it also changed OpenAI.
OpenAI expanded its business everywhere, stretching its front lines longer and longer: ChatGPT serves consumers, Codex advances into the programming field, enterprise agents enter offices, advertising monetizes the attention of free users, Atlas attempts to take over the browser space, and Sora tries to break into the video industry.
There are "princes" everywhere, but none of them is the real "Hong King". Sora stopped its service in April 2026, and the independent browser Atlas exited the market in August. OpenAI began to cut its "side quests" and focus on its core models, Codex, and enterprise products.
What OpenAI actually shares the most similarities with the Taiping Heavenly Kingdom is a trajectory.
An organization gains legitimacy through a grand mission; after its resources and scale expand rapidly, different people begin to fight for the right to interpret the mission; the nominal leader, the technical authority, and the executive group eventually fall into conflict; after the conflict ends, power is re-concentrated, and the original doctrine that once united the organization begins to be reinterpreted by new realities.
The mission used to be OpenAI's most powerful weapon. In the end, it also became the most intractable contradiction for the company.
It Is Increasingly Less Like the Original OpenAI
Altman has won every struggle — the 2023 board of directors showdown, the 2024 public opinion battle, the 2025 renegotiation with Microsoft, and Elon Musk's lawsuit in May 2026. But every time he wins, the company loses one more thing worth defending.
Those who left OpenAI have long regrouped outside. In February 2026, TechCrunch counted 18 companies founded by former OpenAI employees, calling the group the "OpenAI Mafia".
Anthropic has become the industry benchmark; Ilya's SSI is valued at 32 billion USD with no products and no revenue, and NVIDIA has invested 5 billion USD in it; Murati's Thinking Machines is in talks for a 400 billion USD valuation; Elon Musk's xAI was acquired by SpaceX and became an important part of the Mars space narrative.
The most eye-catching one among them is Anthropic. Founded by the Amodei siblings after they left OpenAI in 2021, it is now the benchmark that the entire industry is competing against.
When new models were released in the past, people used to ask: How much better is it than GPT? Now more and more people are asking: How does it compare to Claude Fable?
The commercial changes are even more obvious. In May this year, Anthropic's annualized revenue was only about 47 billion USD; by the end of July, it had exceeded 65 billion USD, growing by nearly 40% in more than two months. During the same period, OpenAI's annualized revenue was about 40 billion USD.
Their positions in the capital market are also reversing. On June 1, Anthropic announced that it had confidentially submitted its S-1 form to the U.S. Securities and Exchange Commission, officially launching its IPO process. Seven days later, OpenAI announced that it had also submitted its S-1 form.
Anthropic may go public this autumn, while OpenAI's IPO this year is increasingly unlikely, with its timeline pushed back to 2027.
According to reports, the implied valuation of Anthropic's shares in some secondary transactions has reached the trillion USD level; around its IPO, the market is even discussing a valuation of 2 trillion USD. In contrast, OpenAI's valuation after its latest round of financing in March this year is 852 billion USD.
When OpenAI released its most powerful model today, the way to prove how strong it is has become "it has once again surpassed Anthropic".
Of course, this does not mean that OpenAI has failed. There is no systematic decline in its technology, and its "lag" is more of a lag in business performance. Talent loss is not a unique wound for OpenAI, as Anthropic, Google, and Meta are also losing top talents.
OpenAI's fundamental base is still extremely strong. It has over 1 billion active users, more than 50 million paid individual subscriptions, over 2 million enterprise clients, its revenue doubles every year, it has a valuation of 852 billion USD, its S-1 has been submitted, and it has deep ties with Microsoft and NVIDIA.
However, the tragedy of the Taiping Heavenly Kingdom is not that it was defeated by the Qing army, but that before it was defeated, it had already lost itself — after the Tianjing Incident, it still occupied half of the country, but it was no longer the ideal kingdom of "all land cultivated together, all food shared" that it once claimed to be.
OpenAI is now in a similar position: it still occupies the center of the AI landscape, still has the largest user base, the most abundant capital, and one of the top models. But the position it occupies is no longer the one it promised when it first started.
OpenAI has not lost yet. It is just increasingly less like the original OpenAI. And this, perhaps, is its biggest problem.
This article is from the WeChat official account "Zhengu Research Lab" (ID: zhengulab), author: Lin Nuannuan, editor: Zhang Duo, published with authorization from 36Kr.