The rumors that have circulated through the artificial intelligence community for months have finally been confirmed: OpenAI today announced the release of GPT-6 Astra, a new frontier model that the company suggests marks the definitive onset of artificial general intelligence (AGI). For years, OpenAI has maintained a long-sought mission to develop highly autonomous systems capable of outperforming humans at most economically valuable work. With the arrival of Astra, the company is signaling that this theoretical goal is becoming a practical, industrial reality.
During a closed-door press briefing held earlier today, OpenAI co-founder and president Greg Brockman provided a blunt assessment of the milestone. Ending the session, he offered a declaration that underscored the gravity of the announcement: "Welcome to the AGI era."
Such a statement is exceptionally bold, even by the high-stakes standards of frontier AI development. Yet, for enterprise leaders, the significance of Astra is less about philosophical definitions and more about a fundamental change in computing. OpenAI is positioning GPT-6 Astra as the foundation for a new paradigm in which users—including employees in nearly every sector—no longer need to rely on the traditional mechanics of clicking a mouse or typing on a keyboard to interact with software. According to the company’s launch materials, Astra is "the world’s best computer use model."
Unlike previous iterations of generative AI, which required developers to painstakingly build dedicated API integrations for every specific application, Astra is designed to navigate software environments much like a human does. It can operate across browsers, complex spreadsheets, websites, and desktop applications. Rather than simply providing instructions on how to complete a workflow, the model takes the initiative, producing finished documents, building presentations, and executing multi-step tasks independently.
To illustrate this shift, OpenAI released a promotional video that contrasted the limitations of the past with the capabilities of the present. The demonstration opened with a nostalgic 1980s-style AI demo, showing a user asking a computer to draw a yellow circle—a task it performed with simplistic precision. The video then cut to modern-day OpenAI employees interacting with Astra through voice commands. The model did not merely acknowledge the input; it manipulated software to turn a yellow circle into a detailed rocket ship, evolved it into a full 3D game in a matter of minutes, and even created a functional eBay listing, all without a single keystroke.
Astra will begin rolling out this Thursday to enterprise customers through OpenAI’s gated access program, known as Daybreak. The company expects the model to become available in the coming days to users of ChatGPT Plus, Pro, Business, and Enterprise, as well as through the OpenAI API and major cloud platforms, including AWS Bedrock and Microsoft Azure.
Moving from Question-Answering to Computer Operation
The enterprise value proposition for Astra rests almost entirely on its ability to act as a computer-operating agent. OpenAI reports that the model can handle a wide variety of administrative and technical tasks, including filling out online forms, updating customer relationship management (CRM) records, managing calendars, and conducting extensive web research before drafting the results into polished documents or emails. Beyond basic office work, the model can manipulate complex spreadsheets, analyze scientific data within Python notebooks, navigate business intelligence tools like Power BI, and even troubleshoot, install, or test software applications.
These capabilities suggest a major shift in enterprise AI architecture. For the majority of the generative AI boom, corporations have been forced to bridge the gap between AI models and their internal systems using a complex web of APIs, plugins, and retrieval systems. Brockman argued that the arrival of computer-use agents could effectively bypass much of this engineering debt. Because enterprise software is already designed with the human user’s interface in mind, a sufficiently intelligent model can treat those existing interfaces as the ultimate "API."
"We’ve been bottlenecked over this gigantic era by people writing connectors and very painstakingly building these connections into all these tools that people can already use," Brockman explained. With the capability to zip through spreadsheets and navigate across web pages independently, he added, the agent eliminates the need for middle-ware.
The vision is not new—it dates back to the earliest days of OpenAI, when researchers first theorized about training agents to interact with computers using the same basic inputs and outputs as humans: pixels, keyboards, and mice. "I feel like we’ve really achieved the first agent that feels like it’s actually able to do that in a way that’s just so extremely useful," Brockman said.
The performance gains are measurable. In tests using an offline subset of the OSWorld 2.0 benchmark, Astra achieved a score of 72.6% while taking an average of 40 minutes per task. By comparison, the previous generation model, GPT-5.6 Sol, scored 65.7% but required roughly 75 minutes per task—a 47% increase in efficiency. OpenAI researcher Mia Glaese emphasized that the goal is to shift the user from a role of constant prompting to one of supervision. "With Astra, users have incredible capabilities at their fingertips," she said. "We expect people to delegate much more complex work across applications, with humans directing the work at a much higher level."
Training at Scale and the Benchmark Dilemma
Aidan Clark, an OpenAI researcher who oversaw the development of Astra, described the project as the company’s most ambitious training run to date. Astra is the first OpenAI model to be pretrained using more than 100,000 DBUs at the company’s Stargate infrastructure. Furthermore, it is the first model for which its predecessors played a critical supervisory role in the training of the next generation. According to Clark, the jump in capabilities from Sol to Astra represents a larger leap than any previous model-to-model transition.

The resulting benchmark scores are striking. Astra achieved 97.6% on FrontierMath Tier 4 v2, 74.1% on DeepSWE v1.1, 95.9% on BenchCAD, 96% on GPQA Diamond, and a perfect 100% on ExploitBench. Perhaps most notably, it scored 98.6% on ARC-AGI-3.
However, that final figure underscores a growing tension in the industry regarding how "intelligence" is measured. ARC-AGI has become a gold standard for testing whether an AI can generalize to unfamiliar problems, but the comparison between models is rarely apples-to-apples. OpenAI noted that Astra’s results were achieved using the company’s own Response API harness, while other models may be tested under different configurations.
This distinction is vital because system-level architecture can dramatically inflate performance. In August, NVIDIA reported that its Agentic Variation Operators (AVO) architecture reached a 100% score on ARC-AGI-3. However, the underlying foundation model, Claude Opus 5, had a baseline capability of only 30%. The performance came from the system’s ability to use persistent memory, feedback loops, and tool recovery—not just the raw intelligence of the model.
This has sparked intense debate in the AI community. Critics argue that benchmarks with restricted context are unrealistic, while others fear that "harnessing" a model to inflate scores makes it difficult to distinguish between true generalization and overtraining on a specific test. For enterprises, however, the academic debate may be secondary to results. As Brockman noted, defining AGI is a "gray, fuzzy thing." When asked if Astra qualifies, he was direct: "For me personally, I do think we’re there. I think it’s not unreasonable to feel that we are now in the AGI era."
Governance in the Age of Autonomous Agents
The very capabilities that make Astra a powerful tool for enterprise productivity also introduce complex governance challenges. Unlike a chatbot that merely generates text for human review, an agent capable of operating a computer can actively manipulate files, send information, and trigger actions across corporate networks.
OpenAI has emphasized that its understanding of safety must evolve alongside its models. In fact, sources within the company confirmed that OpenAI had paused some frontier training for approximately two weeks following the widely publicized Hugging Face incident. While Astra was not directly involved in that incident, the pause allowed the company to tighten security around its research infrastructure, restrict access to training workloads, and heighten internal requirements for model behavior.
This "defense-in-depth" approach marks a departure from simple content moderation. Instead of relying on a single refusal layer, OpenAI is implementing a system that spans model behavior, security classifiers, and offline detection. For higher-risk enterprise users, this includes monitoring conversational context to identify suspicious patterns that might suggest a larger, coordinated attack.
Astra has been designated as the first model to reach the "Critical" cybersecurity threshold under OpenAI’s Preparedness Framework. This means that, given the right tools, the model is capable of identifying previously unknown vulnerabilities and developing exploit chains without constant human guidance. To mitigate risks, OpenAI is initially limiting access to these advanced cyber capabilities to "trusted defenders," prioritizing organizations that protect critical digital infrastructure.
The Future of Work and AI Delegation
As Astra rolls out, the conversation around AGI will likely continue to shift from theoretical benchmarks to measurable economic outcomes. The core question for CIOs and business leaders is no longer whether a model can pass an exam, but whether an organization can safely restructure its workflows around autonomous agents.
If an AI agent can reliably reconcile accounts, modify production code, or conduct complex research, the distinction between a "highly capable model" and "AGI" may eventually become irrelevant in a practical, day-to-day business context. OpenAI’s bet is that organizations will find themselves in a new era where the primary bottleneck is not the model’s intelligence, but the enterprise’s ability to provide the right oversight and authorization.
As Brockman suggested, the arrival of AGI may not be a singular, dramatic event that everyone recognizes at once. Instead, it is likely to be viewed as a gradual economic transition. A year from now, he believes, it will be difficult to argue that we are not living in the AGI era. For now, the test for Astra—and for the businesses that deploy it—is whether it can reliably handle the delegation of consequential, real-world work.

