What AGI means and whether OpenAI's GPT-6 Astra shows it has arrived

Left · 2OpenAI president Greg Brockman recently suggested that the company's advanced model, GPT-6 Astra, could mark the arrival of artificial general intelligence (AGI). 1 2 Left · 2OpenAI defines AGI as an AI system that outperforms humans at most economically valuable work, meaning most digital and cognitive tasks such as writing code or managing projects. For now this excludes physical labour such as agriculture or construction, though robotics may change that. 1 2 Left · 2Others define it differently: Demis Hassabis, founder of Google DeepMind, has said AGI should be able to do "pretty much any cognitive task that humans can do", and Yann LeCun, former chief AI scientist at Meta, believes the term is flawed and should be replaced by "human-level AI". 1 2

Left · 2VentureBeat reported that at a closed press briefing in September Brockman said no well-defined moment that everyone would recognise had played out, but added: "For me personally, I do think we're there." 1 2 Left · 1A few days later, Anthropic chief executive Dario Amodei called in an essay, "We must pace the frontier", for the AI industry to slow its development; Sam Altman and Elon Musk signalled agreement within hours. 1 Left · 2The article's author concludes that the evidence points in more than one direction and is inconclusive. The clearest change with Astra is in use: it is built for longer jobs where the user sets a goal and boundaries and the model works out the sequence itself, which OpenAI presents as moving past a human supplying every next instruction. 1 2

Left · 2Astra scored 99.9% on ARC-AGI-3, a test of AI agents in unfamiliar interactive games, against an OpenAI estimate of 48% for the average human. ARC Prize, which runs the benchmark, reported Astra's best score through its shared harness was 62.7%; the 99.9% came from a harness OpenAI built that preserves the model's reasoning between requests. 1 2 Left · 2Independent evaluators differ on how big an advance Astra is. Epoch AI places it at the frontier alongside Claude Fable 5.1, Kimi K3 and GPT-5.5 Pro, while Artificial Analysis did not rate it as radically ahead of GPT-5.6 Sol. Toby Walsh of the University of New South Wales called frontier intelligence "still today very jagged". 1 2 Left · 2On 8 September 2026 OpenAI published a claimed solution to the Navier-Stokes problem, one of seven Millennium Prize problems. Though often reported as Astra's work, OpenAI says it came from an unreleased, more capable internal model running about 10,000 agents for about 88 hours; Astra handled formalisation and verification over a further 17 hours. The mathematical community has not yet checked the proof. 1 2

Left · 2Amodei wrote that within six to 12 months a swarm of agents could take over the internet with a persistent botnet, possibly causing damage of hundreds of billions of dollars. Anthropic's September threat report describes misuse including a group in northern Yemen using Claude code to build guidance software for rockets and a missile, and Russian and Chinese operators automating espionage; in each case a human chose targets. 1 2 Left · 2The author argues this is a narrow capability applied at speed and scale, not general intelligence, that capability and general intelligence are separate things, and that intelligence is increasingly a property of a system of model, memory, tools and agents rather than of a model tested alone. 1 2