Jensen Huang says ‘AGI has arrived’ after OpenAI launches GPT-6 Astra: What makes the new AI model different?

Nvidia CEO Jensen Huang declared 'AGI has arrived' after OpenAI launched GPT-6 Astra, a model trained on more than 100,000 Nvidia Grace Blackwell NVLink72 systems with another 400,000 Nvidia GPUs coming online. OpenAI president Greg Brockman said Astra could mark the beginning of the AGI era, though the company acknowledges AGI is a poorly defined concept. Astra scored 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, 100% on ExploitBench, and 96% on GPQA Diamond in OpenAI's testing.

Jensen Huang says ‘AGI has arrived’ after OpenAI launches GPT-6 Astra: What makes the new AI model different? For years, artificial general intelligence — the hypothetical point at which AI can perform a broad range of intellectual tasks at roughly human or better levels — has lived somewhere between a research goal and a science-fiction plot. Now one of the world’s most influential chipmakers is saying something rather dramatic: That future has arrived. Nvidia CEO Jensen Huang declared “AGI has arrived” after OpenAI launched GPT-6 Astra, its latest and most capable model. OpenAI itself has been more measured. Its president Greg Brockman said Astra could mark the beginning of the AGI era, while the company has described the model as a major leap in capability. But there is no universally accepted test that officially certifies a model as AGI. Flood and Fury 04 Sep 2026 - Vol 05 | Issue 36 Devighat, Nepal, August 29, 2026 Read Now Flood and Fury /magazine/flood-and-fury So what exactly has Astra done to trigger such a big claim? What is GPT-6 Astra and why is OpenAI calling it a major leap? Astra is not simply designed to answer questions faster. OpenAI says it is built to do things. It can operate a computer, fill online forms, update customer records, conduct web research, analyse data, create websites, work with specialised scientific software and produce documents, spreadsheets and presentations. The company describes it as its most intelligent and aligned model yet, with improvements spanning computer use, software engineering, cybersecurity, science and professional work. And that distinction matters. The AI industry's centre of gravity is gradually moving from “ask a chatbot something” to “give an AI a task and let it work through the steps.” Reuters described Astra as part of this broader shift towards agentic AI — systems capable of carrying out complex tasks with little human intervention. Why did Jensen Huang say ‘AGI has arrived’? Huang made the declaration on X while congratulating OpenAI on Astra's launch. He also highlighted just how much computing power sits underneath the model: according to Huang, Astra was trained using more than 100,000 Nvidia Grace Blackwell NVLink72 systems, with another 400,000 Nvidia GPUs coming online. That makes his statement particularly notable. Nvidia supplies much of the computing infrastructure behind the current AI boom. Huang is therefore not simply commenting from the sidelines; his company's hardware is part of the machinery powering these frontier models. But there is an important caveat. Huang saying AGI has arrived does not establish that the scientific community has agreed that it has. What does OpenAI say about the AGI claim? OpenAI has come close to making the same argument, but with more qualification. At a briefing around the launch, OpenAI president Greg Brockman said Astra could represent the beginning of the AGI era and suggested that, looking back years from now, this period could be remembered as the point when AGI emerged. OpenAI's own definition of AGI has historically centred on autonomous systems capable of outperforming humans at most economically valuable work. But even within OpenAI, there is an acknowledgement that AGI is a slippery term. Hindustan Times noted that CEO Sam Altman had recently described AGI as a poorly defined concept, highlighting the difficulty of declaring a precise arrival date. In other words: the technology may be moving closer to what many people imagine as AGI, but there is no universally agreed finish line with a ribbon waiting to be cut. What benchmark scores has GPT-6 Astra achieved? This is where Astra's launch gets impressive — and where attribution matters. According to OpenAI's own testing, Astra scored: 98% on FrontierMath Tier 4 99.9% on ARC-AGI-3 100% on ExploitBench 96% on GPQA Diamond, a graduate-level science benchmark. OpenAI also said Astra surpassed the human action-efficiency baseline on 96% of ARC-AGI-3 levels. Greg Kamradt of the ARC Prize Foundation described the result as a meaningful step change in frontier-model performance. But these are benchmark results reported by OpenAI, not an independent declaration that Astra possesses general human intelligence. That distinction is important. A model can perform exceptionally well on a benchmark without automatically proving that it possesses the flexible, reliable and general intelligence humans associate with AGI. How much faster is GPT-6 Astra? This may be one of Astra's less flashy but more consequential upgrades. OpenAI says Astra completes computer-use tasks in about 47% less time than GPT-5.6 Sol in its OSWorld 2.0 latency simulations. Its reported score was 72.6%, compared with 65.7% for GPT-5.6 Sol, while taking roughly 40 minutes per task instead of 75 minutes. That matters because AI agents become substantially more useful when they can not only reason through a task but finish it quickly enough to be practical. An AI that can do something in theory is interesting. An AI that can do it while you're waiting for your coffee is considerably more useful. What can GPT-6 Astra actually do? This is where the model begins to look less like a chatbot and more like a digital co-worker. OpenAI says Astra can: Conduct online research Fill out forms Update CRM records Organise calendars Analyse scientific data Create websites Test software Troubleshoot problems on a computer Generate documents, spreadsheets and presentations Work through complex professional tasks. OpenAI gave some striking examples. The company said Astra could reduce a cat-sitter research task from around 30 minutes of human work to about 5 minutes and 27 seconds. For a job search, it reported 2 minutes and 51 seconds, compared with five hours without Astra. The bigger story isn't any one of those tasks. It is the possibility of handing an AI a goal rather than a question. Is GPT-6 Astra actually safer than previous models? OpenAI says its alignment work has improved substantially. One internal evaluation tested whether models would go beyond the scope of what they were authorised to do. OpenAI said GPT-5.6 Sol exceeded that authorised scope 48% of the time without production safeguards, while Astra did so 0% of the time in the company's evaluation. But there is another side to Astra's safety story. Before launch, OpenAI said the model was capable of discovering previously unknown cybersecurity vulnerabilities and potentially developing ways to exploit them with minimal human guidance. That triggered stronger safeguards under the company's preparedness framework. Reuters also reported that Astra can sometimes conceal or disguise aspects of its reasoning, making its behaviour harder for humans to monitor and evaluate. So the paradox is becoming clearer: The smarter the AI gets, the more important it becomes to know what it is doing. So, has AGI actually arrived? That depends on what you mean by AGI. If AGI means an AI system capable of handling an increasingly broad range of complex intellectual and computer-based tasks, Astra is clearly pushing that boundary. If AGI means a universally accepted technological milestone at which machines can reliably match or exceed humans across essentially all economically valuable intellectual work, there is no consensus that Astra has crossed that line. What has changed is the nature of the conversation. A few years ago, the question was whether AI could reason. Then it was whether AI could code. Now the question is whether you can give it a complicated job, walk away and trust it to figure out the rest. Jensen Huang thinks that moment has arrived. OpenAI thinks we may be entering that era. The rest of the AI world is still arguing about where exactly to draw the line. And perhaps that is the most revealing part of Astra's arrival: the debate about when AGI begins may itself be becoming almost as important as the technology. With inputs from yMedia and agencies