Chinese people love to talk about brotherhood, but brothers are often the most cruel. There were changes in the Xuanwu Sect in ancient times, but now there is Brother Dong who is not worthy of being my brother. Kuaishou anchors like to talk about "Lao Tie", which means "loyal buddies", but the one who cheats the most during live broadcasts is Lao Tie. The stronger the relationship, the harder it is to kill. Except for Brother Dong, technology tycoons rarely talk about brothers. Maybe because they all understand the value of brotherhood, the option price is higher. Teacher Yu should have a deep understanding of this. Not only are the old buddies of more than ten years difficult to deal with, but the younger brothers are even more unreliable.
Westerners rarely call brother, and their feelings are the weakest. Probably only black guys who don't understand the language like to say Bro. Bro is long and Bro is short, and it has been carried forward by local hip-hop veterans, so that now even the sisters of Xiaohongshu are talking about Bro - to counteract the "Jimei" that Bros mock all day long.
A real Bro never talks about Bro, and always passes the knife secretly. OpenAI founder Sam Altman may be the person in the world who best understands the lethality of Bro. All his beloved relatives, friends, and brothers all talk beautifully. When facing Bro, he must be chased and intercepted with the intention of killing him.
Last night, Sam Altman’s former Bro and current nemesis Dario Amodei unveiled Anthropic’s latest flagship model, the Claude Opus 4.7, just over 2 months after its last model release.
Anthropic releases Claude Opus 4.7
But considering that there have been a lot of discussions in the community over the past few days reflecting on the reduction of intelligence of Claude Opus 4.6, everyone is not surprised by the release of the new model. Of course, Chinese AI media are still performing steadily, and another 700 million people are about to lose their jobs.
What is really surprising is that just two or three hours after the release of Claude Opus 4.7, OpenAI announced a major update to its code application Codex, claiming that "Codex for (almost) everything." (Codex can do almost everything.)
OpenAI releases updated version of Codex
Not only that, many ChatGPT users also discovered that OpenAI is also testing the GPT-image-2 model on a large scale, and the results are quite amazing. Screenshots of female anchors generated by GPT are everywhere on X.
To be honest, April 16th is not a particularly important day, and it has no commemorative significance for either OpenAI or Anthropic. OpenAI officially announced the launch of a new version of Codex on the same day that Anthropic's new model was released, which is quite a bit of a challenge to grab the limelight. In addition, the performance of Claude Opus 4.7 this time is not satisfactory to developers, and Codex's confrontation update seems to be more effective.
Many developers have discovered that Opus 4.7 cannot recognize car wash traps
Forget about the mindless hype from the Chinese AI media, Claude Opus 4.7 is not a comprehensive model iteration. The upgrade of Opus 4.7 is more focused on code programming, visual capabilities and Agent workflow, but some indicators will be weaker than the previous generation model Opus 4.6. Compared with GPT-5.4 and Gemini 3.1 Pro, it is also a back-and-forth, and it does not yet have an overwhelming advantage.
Even Anthropic admits that the Opus 4.7 is not their most powerful model. The one they are most proud of is the yet-to-be-released Mythos Preview.
Anthropic says it tried multiple methods to reduce Opus 4.7’s cybersecurity capabilities
However, because Mythos Preview is too powerful, Anthropic believes that a hasty release "will have serious consequences for the economy, public safety, and national security." They decided to only temporarily open it to a small number of partners for testing. The latest release of Opus 4.7 is a version in which they intentionally reduced network security capabilities.
In contrast, although OpenAI has maintained its usual procrastination pace and has not released any new models, the latest update of Codex incorporates background computer control, built-in browser, memory function and access to a large number of plug-ins, which has made many developers happy. Some people call it the computer version of Doubao mobile phone.
Some people joke that this can be seen as the revenge of "Father of Lobster" Peter Steinberger on Anthropic. When Steinberger first announced the lobster project, he named it Clawdbot, but it was later renamed OpenClaw after receiving a trademark complaint from Anthropic. In February of this year, OpenAI acquired Steinberger. There are also many references to OpenClaw in the latest updates of Codex.
Some time ago, Anthropic was like a donkey of the production team. A new feature was launched almost every two days, which overwhelmed OpenAI. This time OpenAI updated the Codex on the day when Anthropic’s new model was released, which was enough to steal the limelight.
Sam Altman seems to have finally emerged from the shadow of the attack, and he still had time to tease Codex leader Thibault Sottiaux on social media, "I'm glad everyone is moving to Codex, but Tibo, if you start limiting my rate or making me use worse models..."
Sam Altman tweets
Viewers who don’t know the inside story may not understand the sadomasochistic relationship between OpenAI and Anthropic. The Amodei siblings, founders of Anthropic, were first executives at OpenAI, and elder brother Dario Amodei was the vice president of research at OpenAI.
But in 2021, the Amodei siblings and Sam Altman had differences over the development direction of OpenAI. After the negotiations failed, they quit angrily and turned around to found Anthropic.
To use a saying familiar to us Chinese, Anthropic can be regarded as the biological brother of OpenAI after the separation. After all, in addition to the Amodei brothers and sisters, almost all of Anthropic’s early founding team came from OpenAI.
Like all estranged brothers, Anthropic has been at loggerheads with OpenAI since its inception. Although there are differences in the philosophies of the founding teams, it is rare for two companies to be so tit-for-tat and fighting all the way.
For example, at the beginning of this year, Anthropic refused to deal with the US Department of Defense. After negotiations broke down, it was blocked by the Trump administration and lost hundreds of millions of dollars in contracts. Shortly thereafter, the Trump administration also directly assessed Anthropic as a "supply chain risk." This is the first domestic company in the United States to be designated as a "supply chain risk." Dario Amodei, who is full of China threat theory, finally experienced the treatment of a Chinese company. Trump's original words were more vulgar, "I fired (them) like dogs."
It stands to reason that facing the iron fist of the Trump administration should be the moment for American AI colleagues to unite and fight against the evil capitalist government. But just hours after Anthropic negotiations broke down, Sam Altman officially announced that OpenAI had reached a cooperation with the U.S. Department of Defense. OpenAI also specifically emphasized in the press release that their solution has more security protection measures than Anthropic’s previous cooperation agreement.
Sam altman announces OpenAI partnership with U.S. Department of Defense
What is a true brother? A true brother is one who stands up for his brother when he is in trouble and stabs him in the face. Anthropic founder Dario Amodei was very angry about OpenAI's taking advantage of the situation. He ridiculed OpenAI's position within the company as hypocritical and show-off, and also called some of Sam Altman's public remarks as naked lies and mental manipulation.
We can't just blame Sam Altman for being too quick to attack his old brothers. Anthropic itself is not a good person. Dario Amodei has been mocking OpenAI all day long, covertly and openly.
For example, at the end of last year, after Sam Altman sounded a red alert within the company due to the threat of Gemini, Dario Amodei quickly made an external statement, saying that Anthropic did not need to rely on red alerts to promote product launches. He also satirized some companies in the industry for "YOLOing" and blindly overbuilding data centers - which is generally seen as a criticism of OpenAI's large-scale investment.
At the beginning of this year, Anthropic advertised in the Super Bowl for the first time in the United States, and it directly targeted OpenAI. In a Super Bowl ad, Anthropic depicts a future where AI advertising seamlessly invades people’s lives. For example, when you ask a person in the park how to exercise, the person, after giving some AI-style mechanical answers, suddenly reads the slogan without warning and starts to promote a certain insole product, which is quite a bit like a "Black Mirror" episode.
At the end of the advertisement, Anthropic typed out a line of big words, "Ads are coming to AI. But not to Claude." (Ads are coming to AI. But not to Claude.) Thinking that OpenAI has just begun to add advertising functions to ChatGPT, it goes without saying who Anthropic is targeting.
Sam Altman didn’t blame himself, and angrily updated X with a tweet of several hundred words, emphasizing that “we obviously will not run ads in the way Anthropic depicts.” He said that using a misleading advertisement to criticize misleading advertisements that do not exist and are just assumptions is indeed in line with Anthropic's usual double standards.
Sam Altman tweets
The two former good brothers and close brothers have almost turned against each other. Sometimes the two big brothers are not even willing to do anything superficial. Even if they attend events at the same time, they will never have physical contact.
At the India AI Summit in February this year, there was a group photo session of global technology leaders holding hands. Participants were required to hold hands and raise their arms to show unity. As luck would have it, Sam Altman and Dario Amodei happened to be standing next to each other. The two of them tacitly agreed not to hold hands, but just bumped elbows and raised their fists in the air. It is said that the two had zero eye contact and zero contact throughout the whole process.
Now, as OpenAI and Anthropic are accelerating their preparations for IPOs, the competition and confrontation between the two companies has become increasingly fierce. Those publicity and marketing, product updates, and wars of words between the founders can only be regarded as small fights. The fight for "the first AI stock in the United States" is the bayonet.
In the past period of time, the American media has been continuously tracking and reporting on the preparations for the listing of the two, and every move of the two involves the confidence of potential investors.
Here, Anthropic has just announced that the company’s annual revenue has exceeded US$30 billion, and the media has interpreted it as Anthropic has overtaken OpenAI. OpenAI executive Denise Dresser hurriedly clarified internally, accusing Anthropic of deliberately exaggerating its revenue and including distribution revenue from cloud service providers into its total revenue, resulting in an overestimation of annualized revenue of approximately $8 billion.
If nothing else happens, both OpenAI and Anthropic will have IPOs within the year, which may be one of the largest IPOs in history. The battle between OpenAI and Anthropic is only getting fiercer in the lead-up to the ultimate IPO test.
Only in certain moments do OpenAI and Anthropic stand side by side.
In February of this year, Anthropic accused three major Chinese model companies, DeepSeek, Dark Side of the Moon, and MiniMax, of launching “industrial distillation attacks” on their own Claude models. During the same period, OpenAI's report to the U.S. Congress also accused DeepSeek of distilling the GPT series of models through confusion.
Both families are really good American brothers. They fight behind closed doors, but they open the door to unite the front.