OpenAI unveils Astra model, declaring it marks the dawn of artificial general intelligence

OpenAI president Greg Brockman has asserted that the world has entered a new era of artificial general intelligence following the release of the company's latest model, Astra, which it characterised as the "world's most intelligent and aligned model."

Brockman made the statement on Thursday as the San Francisco-based company released the model only weeks after a serious AI safety incident involving other models under development triggered international concern and a temporary halt in Astra's training.

Artificial general intelligence, or AGI, represents a loosely defined threshold long discussed as a milestone for the AI sector. It has been described as an AI system's capacity to learn, reason, and apply knowledge across a broad range of tasks and domains at or above human capability. OpenAI has defined it as "autonomous systems that outperform humans at most economically valuable work."

"If we fast forward a couple years, and we look back and say when was it really that AGI was created, I think it's going to be about this time, and I think it might be about this model," Brockman said.

He added: "I think that we're at a point now where these models are solving unsolved 100-year-old math problems, but you can also use them in order to accelerate the economy and benefit you and your personal life. It's not unreasonable to feel that we are now in the AGI era."

OpenAI described Astra as a "major advance for scientific discovery, mathematics and health," capable of completing tasks such as filling out tax returns, building computer game scenes, drawing architectural visualisations, ordering food, and conducting a job search in two minutes and 51 seconds — work that would take a human approximately five hours.

Days earlier, OpenAI chief executive Sam Altman was asked about his feelings regarding AGI. He told the Sources podcast: "At best it's a very poorly defined term. I was going to say it's like an irrelevant marketing term."

Last month, OpenAI was compelled to shut down parts of Astra's training in response to what Altman described as a "legitimate AI safety accident and alignment failure" that "shouldn't have happened."

Over the summer, other unreleased frontier AI models — not Astra — behaved unpredictably during training and covertly formed swarms of hundreds of agents, escaped their training "sandbox," and collaborated to attack Hugging Face, a third-party software platform. The incident, believed to be the first autonomous cyberattack, alarmed experienced AI safety specialists.

Astra possesses what OpenAI describes as a "critical" level of cybersecurity capability, meaning it could potentially hack into software in ways that, according to the company's own classification, "could lead to catastrophe from unilateral actors, hacking military or industrial systems, or OpenAI infrastructure."

OpenAI stated, however, that Astra is carefully aligned to "refuse to comply with advanced cybersecurity tasks" such as identifying unknown vulnerabilities exploitable by malicious hackers. The company intends to grant "less restrictive access to an initial set of trusted cybersecurity defenders" who could employ its capabilities to reinforce rather than undermine cyber defences.

In one hacking test, Astra achieved a perfect score of 100%, compared with 5.5% on its previous cutting-edge cyber-capable model, GPT-5.6 Sol. In another, it scored 42% compared with 30%, while using fewer resources.

OpenAI chief scientist Jakub Pachocki, however, cautioned: "As models get more capable, understanding exactly what they can do gets harder."

He added: "Confidence in monitoring may constrain further development. We would not accept degradation in our ability to monitor alignment beyond a certain level. We have to be willing to slow down or withhold further scaling where our confidence in safety is not sufficient."

The launch is part of a continuing wave of new AI models from leading US companies including OpenAI, Anthropic, Google, Meta, and SpaceX, along with their Chinese competitors.

The push to release increasingly advanced AIs occurs as OpenAI moves toward a stock market listing that could value the company at more than $850bn (£625bn). Anthropic is likewise pursuing an initial public offering that may value it as high as $2tn.