OpenAI’s president, Greg Brockman, currently suggested that its advanced model, called GPT-6 Astra, could mark the arrival of something known as artificial general intelligence, or AGI.
If AI itself is often described as the ability to simulate human intelligence, OpenAI describes AGI as an AI system that outperforms humans at most economically valuable work—to be better at the majority of digital and cognitive tasks a human can perform. This includes task such as writing computer code, organizational activities or managing projects.
For now, at least, it excludes work that includes physical labor, such as agriculture or construction. The incorporation of AI with robotics, even though, means such sectors of industry might not be the exclusive domain of humans forever.
This definition by OpenAI targets on breadth of competency by an AI system. Moreover, other experts keep that AGI will be characterised by additional cognitive leaps beyond present-day AI models. For example, Demis Hassabis, founder of Google DeepMind, has stated that AGI should be able to do “pretty much any cognitive task that human can do.”
This definition makes AGI more about brilliance and a bit less about depth. Yann LeCun, former chief AI scientist at Meta, believes the term AGI itself is flawed and should be replaced by “human-level AI.”
These signifies snapshots from a vigorous ongoing debate over when AGI will arrive and how to recognize it.
Venturebeat magazine reported Brockman acknowledging this at a closed press briefing about GPT-6 Astra in September when he said, “When we begun OpenAI, we kind of thought that there was going to be this well-defined moment that everyone might recognize … it hasn’t played out like that.”
But he added, “For me personally, I do think we are there.” Just a few days after Brockman made his comments, the chief executive of Anthropic, Dario Amodei, known as in an essay titled “We must pace the frontier” for the AI industry to gradual down the pace of its development. Within hours of this essay being published, both Sam Altman and Elon Musk signaled their agreement.
So is Brockman right about AGI being here? The evidence still points in more than one direction and is therefore inconclusive.
The clearest shift with GPT-6 Astra is in how people use it. Earlier models require detailed instructions at every step. Astra is built for longer jobs wherein the user sets a aim and some limitations, and the model works out the sequence itself.
OpenAI’s demonstrations showed the frontier model managing numerous unrelated task at once. The company gives this as moving past the pattern wherein a human supplies the next instruction every time .
That is a real change, and it is the part most user will observe first.
One of the benchmarks indicates that Astra scored 99.9% on ARC-AGI-3, a test that locations an AI agent in strange interactive game with no instructions and measures how efficiently it works out the rules.
OpenAI estimates that the average human scored 48% when asked to workout the same unexpected games. The detail matters here, moreover. ARC Prize, which runs the benchmark, said that Astra’s best score via the shared “harness” it offers every model was 62.7%. A harness is a type of software operating system, or physical environment, inside which the AI system works. ARC Prize offers Astra with this software harness.
The 99.9% came from a different harness, constructed via OpenAI for Astra, which preserves the AI model’s reasoning between requests. Importantly, Astra did better when it was capable of reuse previous work and wasn’t “beginning from scratch” between requests in terms of its reasoning.
Evaluators disagree
Independent evaluators disagree about how much of an strengthen GPT-6 Astra signifies on its predecessor. The independent research institute Epoch AI places it right on the frontier, along different cutting-edge AI models which include Claude Fable 5.1, Kimi K3 and GPT-5.5 Pro. The independent evaluation company Artificial Analysis did not rate it as being radically ahead of its predecessor—GPT-5.6 Sol.
Toby Walsh, an AI researcher on the University of New South Wales in Sydney, Australia, defined the intelligence in frontier AI models including GPT-6 Astra and others as “nevertheless today very jagged,” telling Al Jazeera that there are simple things that even the best AI models do poorly.
On Sept. 8, 2026, OpenAI published a claimed solution to the Navier-Stokes issue, considered one of seven Millennium Prize issues in mathematics. This has often been pronounced as Astra’s work. It was not. OpenAI says the proof came from an unreleased internal model, significantly more capable than Astra, running around 10,000 AI agents (AI models that perform tasks in response to human instructions) in parallel for about 88 hours.
Astra’s contribution was the formalization and verification afterward, which took a further 17 hours. The proof has not yet been checked via the mathematical community.
Amodei’s September essay calling for an AI slowdown stays the most strongest current statement about the risks of frontier models. He wrote that within 6 to 12 months, a swarm of agents could be capable of taking over the internet with a persistent botnet. This ought to lead to the catastrophic failure of key systems and would cause damage possibly running into hundreds of billions of dollars.
Claude misuse
It is worth inspecting Amodei’s warnings. Anthropic’s September threat report sets out examples of how AI is already being utilized in alarming ways.
As per the Anthropic report, a group in northern Yemen lately used Claude Code to form guidance software for rockets and a missile. Russian and Chinese operators automatic espionage work. A single French-speaking individual used AI to build a software platform that made personal information about people affiliated with a aimed political movement searchable online. In each case, a human chose the goals and the model supplied the labor.
That is a narrow capability applied at speed and scale under human path. It is serious, but it isn’t a description of general intelligence.
Capability and general intelligence are separate things, and they call for one of a different responses. Amodei’s proposed response is to slow frontier development at the same time as expanding independent evaluation and coordinating safety standards across nations. A absolutely general system would raise questions that those measures are not designed to reply.
GPT-6 Astra does not settle whether AGI has arrived. Rather than, it exposes why that question is turning into more difficult to answer. Intelligence is increasingly a property of an AI system comprising a model, memory, tools and agents, rather than of a model tested in isolation.
On the narrower question of whether these systems can now cause extreme harm, the people running the companies have said plainly that they actually can.
