The company behind ChatGPT says “the AGI era has arrived” — here’s why so many people are raising an eyebrow
🎧 Post Summary
📌 Post Summary
OpenAI released a new AI model called “GPT-6 Astra” on September 3 and declared that “the AGI (artificial general intelligence) era has arrived.” Nvidia CEO Jensen Huang backed the claim too. Astra’s core pitch is that it can do anything a human can do on a computer — it reportedly finished a search that normally takes 30 minutes in 5 minutes, and a job-search task that takes 5 hours in 3 minutes. But on an independent intelligence benchmark, Astra actually scored lower than a rival model, and AI researcher Gary Marcus called the announcement “a victory lap with no evidence behind it.” At the heart of the debate: there’s still no agreed-upon definition of what “AGI” even means.
“AGI has arrived” — you’ve probably seen that phrase floating around lately. OpenAI said it when launching a new AI model, and now experts are arguing over whether it’s actually true. Here’s a plain-language rundown of what AGI means and why this is such a big deal.
① So what exactly is AGI?
AGI stands for Artificial General Intelligence. In plain terms, it’s not “an AI that’s great at one thing” — it’s “an AI that can handle pretty much any intellectual task a human can.” Models like ChatGPT have been great at individual skills — writing, coding, drawing — but they couldn’t independently judge and juggle a whole range of tasks the way a person can. OpenAI is claiming its new model, GPT-6 Astra, has crossed that line.
② What did it actually show off?
OpenAI’s demo videos are what got people talking. A pet-sitting search that takes a human 30 minutes was finished by Astra in 5 minutes and 27 seconds. A job-search task that normally takes 5 hours was done in 2 minutes and 51 seconds. It works by controlling the mouse and keyboard directly — looking at the screen, clicking, and finding information, just like a person would. One developer said he linked Astra up with other tools and built a 3D game demo in 45 minutes. The benchmark scores (tests that measure AI performance) were eye-catching too — Astra reportedly scored a perfect mark across every subject on a Korean college entrance exam-style test.
③ The people saying “AGI has arrived”
OpenAI president Greg Brockman said, “if it’s something a human can do on a computer, it’s reasonable to view Astra as the first AGI.” CEO Sam Altman said he hopes it becomes “a new starting point for startups, scientific discovery, and creativity.” Then Nvidia CEO Jensen Huang jumped in too, declaring “AGI has arrived. Congratulations to the OpenAI team” — which is what really lit the debate on fire. He added that Astra was trained on more than 100,000 Nvidia GPUs, and that it took four years to get from ChatGPT’s earliest model to this point.
④ The pushback: “Not even close”
Here’s the twist: on the independent Artificial Analysis Intelligence Index (AAII), Astra scored 55 — lower than rival Anthropic’s Claude Fable 5.1, which scored 57 — landing it in a tie for 5th place overall. AI researcher and NYU professor emeritus Gary Marcus went after the claim directly, saying it was “a victory declared with no evidence or definition behind it.” He argues Astra meets only 1 or 2 of the 10 criteria he uses to judge whether something is AGI. Notably, OpenAI and Microsoft once had a contract clause tied to whenever AGI was achieved — but they deleted that clause entirely back in April, citing the lack of any agreed-upon definition.
⑤ So does this actually affect me?
For now, free users can’t even try Astra — it’s paid-plan only, and the price has gone up from the previous model. There’s also a notable safety wrinkle: Astra found two previously unknown security vulnerabilities on its own. Because of how capable it turned out to be, it became the first model to hit OpenAI’s highest internal risk rating, “Critical.” In response, OpenAI also rolled out a $1 billion cyber-defense program called “Daybreak.” The more powerful the capability, the more careful the risk management has to be.
📌 Key Takeaways
- The term “AGI” still doesn’t have a single definition that everyone agrees on.
- Benchmark rankings can shift a lot depending on the testing method, so a single score is hard to treat as an absolute measure of performance.
- “This time it’s really different” is a claim that’s come up with pretty much every major AI release — worth keeping in mind.
- With prices rising and no free-plan access, most everyday users may not feel much of a difference just yet.
In the end, this debate is less about “how smart is Astra, really” and more about “how do we even define AGI.” OpenAI naturally wants to say it’s finally arrived — but declaring you’ve reached a goal that has no agreed definition is bound to be controversial on its own. The performance leap looks real. Whether it actually counts as “AGI” is a fight that’s likely to keep going for a while.
References
- Money Today (mt.co.kr) — “”AGI Has Arrived,” Jensen Huang Declares Too… GPT-6 Astra Scores Perfect on Korean CSAT”
- CBC News (cbci.co.kr) — “OpenAI Unveils ‘GPT-6 Astra’… Jensen Huang Says ‘AGI Has Arrived’ vs. Academia’s ‘Baseless Victory Lap’”
- News Space (newsspace.kr) — “[Big Tech Column] The First AI Model to Score Perfect on Korea’s College Entrance Exam… The Question GPT-6 Astra Raises: ‘Has AGI Already Arrived?’”
- Now Pick (wikidocs.net) — “GPT-6 ‘Astra’ Unveiled: OpenAI’s Vision of the Start of the AGI Era”
- Wikipedia (ko.wikipedia.org) — “GPT-6 Astra”
- Namu Wiki (namu.wiki) — “GPT-6”