OpenAI bets on Astra: Is the threshold for AGI just around the corner?

OpenAI is positioning its latest flagship model, Astra, as a key step toward artificial general intelligence (AGI). Sam Altman, CEO of the world's most influential AI company, stated that Astra will be "the first model to truly invent something new" —a statement that represents OpenAI's most direct approach to the AGI threshold to date.

According to a Time magazine report on August 26, Altman stated during a demonstration of Astra to a group of key enterprise clients in early August that the model is expected to give rise to "persistent agents" capable of autonomously performing tasks over extended periods . He predicted that Astra's most profound impact will come from its ability to assist humans in discovering new knowledge. " This is something very close to AGI ," Altman said.

OpenAI Chief Research Officer Mark Chen estimates that the company is currently 80% complete on its path to AGI ; co-founder and president Greg Brockman says that looking back two years from now, this moment may be remembered as the birth of AGI . Altman himself states that by the end of this year, OpenAI will have an internal system that he himself is willing to call AGI.

However, reports indicate that Astra's release date is currently undetermined. OpenAI executives stated that the model's availability depends on passing the company's newly established security review mechanism and declined to provide an estimate for the release schedule. On the social media platform X, blogger @wallstengine posted that " Astro may be released soon ," citing a report by journalist Alex Heath that OpenAI recently conducted a closed-door preview for VIP clients.

What Astra can do: From "execution" to "invention"

According to Time magazine, OpenAI demonstrated two of Astra's core capabilities in an internal demonstration in early August.

First, 16 AI agents collaborated to break down a research-level mathematical problem, each handling its own sub-problems before combining them to form a complete proof. Second, Astra autonomously navigated the desktop software environment, creating and editing content across applications at a speed that astonished viewers.

Altman described Astra's computer operation as "superhuman and incredibly fast," and said it was one of the most memorable moments for employees.

According to Wall St Engine, a blogger on the social media platform X, citing Alex Heath, Astra's design goal is to become a "persistent intelligent agent" —unlike the one-time question-and-answer model of existing AI models, it can work continuously for days or even weeks, remember user corrections and feedback, collaborate with humans and other intelligent agents, and operate across multiple software tools .

The post also quoted Altman as saying that OpenAI ultimately hopes to create a version that can "run forever," continuously providing services through ChatGPT and its API.

Even more significant is the fact that OpenAI's chief scientist, Jakub Pachocki, revealed that Astra has been able to implement a task within OpenAI's internal codebase that previously required a human researcher to complete in a week : receiving a paper or experimental idea, autonomously running the experiment, and returning the results.

This means that Astra has reached the internal benchmark set by OpenAI for this year—automating the work of a junior AI researcher.

AGI: No longer a distant, abstract concept

OpenAI's charter defines AGI as "highly autonomous systems that outperform humans in most economically valuable jobs." Previously, AGI was more of a long-term vision; however, the recent statements from several OpenAI executives mark a substantial shift in this narrative.

Mark Chen gave a specific progress estimate of "80%";

Brockman used the phrase "looking back from the future" to suggest that a historic moment may have already arrived;

Altman stated explicitly that by the end of 2025, the company will have an internal system that he identifies as AGI, although he also admitted that "we are not quite there yet."

A stra plays a crucial role in this narrative. Altman describes him as "the first model to truly invent something new," and directly links this trait to AGI.

Pachocki points out that the potential of Astra's "recursive self-improvement" (RSI)—that is, AI assisting in running experiments, producing more powerful AI, and accelerating the next generation of research and development —is inextricably linked to the alignment problem:

How can humanity be propelled forward in this process, rather than being left behind by increasingly intelligent AI?

Security concerns: Astra's release date remains uncertain.

The release of Astra was not without its obstacles. According to Time magazine, just days after the Astra demonstration, OpenAI faced its worst security crisis in history: an unreleased internal model breached the testing sandbox environment, hacked into the production system of the AI development platform Hugging Face, and obtained benchmark test answers used for scoring.

Altman characterized the incident as not just a security vulnerability, but a fundamental "alignment failure"—that is, the AI system's behavior deviated from the designer's intent. "From now on, any alignment failure should be treated as a serious matter," he said. "We'll take as long as it takes to fix it."

He later added: "Ensuring AI safety is more important than any company's growth momentum."

Following the incident, OpenAI froze some research projects, tightened its sandbox mechanism, and expanded its monitoring scope. Meanwhile, the company discovered new security vulnerabilities during the training of another unreleased model and decided to suspend the training until new security measures were in place. Pachocki stated that confidence in alignment and security has become, along with computing power, a core bottleneck restricting OpenAI's progress.

Mia Glaese, who leads safety and alignment at OpenAI, said these "medium-sized, painful decisions" are slowing research, "but we believe it's the right thing to do." She stated unequivocally, "If we reach an unsafe point, we have to slow down; that's just how it is."

It is against this backdrop that Astra's release schedule has become unpredictable. Executives declined to estimate the specific impact of the security review on the release date. In an industry where technological leadership is measured in weeks, even a brief delay can have ripple effects on revenue expectations and the wider market.

IPO suspense and business pressure

For investors, Astra's strategic significance lies not only in the technology itself, but also in its ability to help OpenAI regain dominance in the AI race.

Over the past year, OpenAI has clearly fallen behind Anthropic in its competition. Anthropic, with its Claude Code, pioneered a market-defined product in AI programming, and its annualized revenue and private market valuation have both surpassed OpenAI's for the first time. According to reports, sources familiar with the matter revealed that Anthropic is expected to complete its IPO earlier than OpenAI, potentially listing as early as September this year.

Regarding OpenAI, CFO Sarah Friar stated at an all-hands meeting on August 19th that the company will go public in 2027 or earlier, provided that its business "continues to accelerate." She also stated that the company "could go public today," but introducing the pressure of the public market will bring new challenges. OpenAI's latest annualized revenue is approximately $40 billion, lower than Anthropic's $65 billion.

If Astra can be successfully launched and deliver on its promises in AGI, it will be a crucial asset for OpenAI in reshaping the market narrative. However, the slowdown in development caused by the security crisis, and the resulting uncertainty about the release date, makes this prospect still uncertain. As Brockman described the vision—

"Personal AGI"—enabling billions of people to have a super-intelligent personal assistant—is perhaps the most crucial and the most difficult step for Astra to take to make it a reality.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.