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. Described by the company as its most significant frontier model to date, Astra is being framed by OpenAI leadership as the long-awaited milestone marking the onset of Artificial General Intelligence (AGI)—the company’s foundational goal of creating highly autonomous systems capable of outperforming humans at the vast majority of economically valuable work.
In a closed-door press briefing held earlier today, OpenAI co-founder and president Greg Brockman provided a remarkably direct assessment of the launch, concluding the session with a declaration that signals a shift in the industry’s trajectory: "Welcome to the AGI era."
Such a bold framing is rare, even within the hyper-competitive landscape of frontier AI development. However, for enterprise leaders and technology executives, the significance of Astra lies less in philosophical labels and more in its immediate, practical utility. OpenAI is positioning GPT-6 Astra as the foundation for a new era of computing, one where the traditional interface of keyboard and mouse becomes optional. According to launch materials provided to the media, OpenAI describes Astra as "the world’s best computer use model."
Unlike previous iterations of generative AI, which required developers to painstakingly build dedicated API integrations for every piece of software an agent needed to interact with, Astra is designed to navigate software much like a human does. The model is capable of operating across web browsers, complex spreadsheets, websites, and native desktop applications. It does not simply guide a user through a process; it produces finished documents, builds presentations, and executes multistep workflows autonomously.
To demonstrate this, the company showcased a promotional video contrasting a 1980s AI demo—where a computer performed the rudimentary task of drawing a yellow circle—with the capabilities of Astra. In the demonstration, OpenAI employees interacted with the model entirely via voice, asking it to transform a yellow circle into a rocket ship, then into a full 3D game, and finally to generate a live listing on eBay, all without manual input. Astra is scheduled to begin rolling out this Thursday to enterprise customers via OpenAI’s "Daybreak" gated access program, with a broader release for ChatGPT Plus, Pro, Business, and Enterprise customers, as well as the OpenAI API and cloud platforms like AWS Bedrock and Microsoft Azure, expected in the coming days.
From Answering Questions to Operating Computers
The primary value proposition for businesses centers on the model’s ability to act as a digital operator. OpenAI reports that Astra can autonomously fill out online forms, update CRM records, organize calendars, conduct sophisticated web research, and draft high-quality documents or emails. Its technical proficiency extends to manipulating spreadsheets, analyzing scientific data within Python notebooks, working in business intelligence tools like Power BI, creating and testing websites, and even troubleshooting software installations.
This evolution suggests a fundamental shift in enterprise AI architecture. Throughout the initial generative AI boom, corporations have been forced to build custom bridges—using APIs, plugins, and complex retrieval systems—to connect models to their proprietary data and tools. Greg Brockman argued during the briefing that computer-use agents like Astra could bypass much of this infrastructure because modern software is already designed to be used by humans. By mimicking the human user, the model leverages the existing interfaces of these applications directly.
"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 noted. With the advent of highly capable computer-use agents, he added, the focus shifts toward models that can "zip through spreadsheets, fill out forms, and navigate across web pages" as a human would.
OpenAI researchers believe this is the realization of a concept discussed during the company’s earliest days: training an agent to interact with the basic inputs and outputs available to humans, namely pixels, keyboards, and mice. According to data shared by the company, on an offline subset of OSWorld 2.0, Astra achieved a score of 72.6% while taking approximately 40 minutes per task. This represents a significant improvement over the previous model, GPT-5.6 Sol, which scored 65.7% while taking roughly 75 minutes per task—a 47% reduction in time.
A Landmark in Training Scale
Aidan Clark, an OpenAI researcher, characterized the development of Astra as the company’s largest-scale training run to date. Astra is notably the first OpenAI model pretrained using more than 100,000 DBUs at the company’s Stargate infrastructure. Furthermore, it is the first model for which previous generations played a significant role in supervising the training of their successor. Clark noted that the leap from Sol to Astra represents a larger increase in capabilities than the jump from previous models to Sol, driven by a combination of large-scale pretraining and reinforcement learning designed to teach the model how to connect disparate information and execute long-horizon tasks.
The benchmark results are impressive, with Astra scoring 97.6% on FrontierMath Tier 4 v2, 95.9% on BenchCAD, and 100% on ExploitBench. It also recorded a 98.6% score on the ARC-AGI-3 benchmark, a test specifically designed to measure an AI’s ability to generalize to unfamiliar problems.
However, the industry’s reliance on such benchmarks remains a point of contention. The 98.6% score on ARC-AGI-3 has ignited a debate regarding what actually constitutes "intelligence" in an AI system. Comparisons are often difficult because performance can be heavily influenced by the "harness" or the system surrounding the model. For instance, NVIDIA recently reported that its Agentic Variation Operators (AVO) architecture reached a 100% score on ARC-AGI-3, but this was achieved by wrapping an underlying model—which had a baseline capability of only 30%—with memory, feedback, and recovery tools.
This debate highlights a growing challenge: how do we define the object being measured? Is it the raw foundation model, the model combined with persistent memory, or the entire deployed system? For enterprises, the distinction is increasingly academic. Companies are concerned with outcomes—reliability, auditability, and cost—rather than the source of the intelligence.
Governance in the Age of Autonomy
The very capabilities that make Astra powerful also introduce complex governance challenges. A standard chatbot generates text for human review; an agent operating a computer can modify records, send data, and take irreversible actions across enterprise applications. Mia Glaese, an OpenAI researcher, emphasized that as users delegate more complex tasks, the need for trust becomes paramount.
The company’s internal safety preparations for Astra were rigorous. OpenAI sources revealed that the company paused some frontier training for two weeks following an incident involving Hugging Face, even though Astra was not involved in that specific event. During this pause, OpenAI tightened security protocols, restricted access to training workloads, and heightened internal requirements for model behavior. This "defense-in-depth" approach includes not only built-in refusals but also system-level classifiers and offline detection tools designed to identify abuse patterns that span multiple prompts.
This is a critical pivot for enterprises: the control surface is no longer just the prompt, but the entire sequence of actions the agent takes. Organizations must now account for how an agent understands its authorization boundaries and whether it can recognize when a request requires it to step outside those limits.
A New Economic Metric
OpenAI is also attempting to change how customers evaluate cost. Greg Brockman noted that traditional token-based pricing is becoming an obsolete proxy for value in an agentic world. Instead, he argued that businesses should focus on "price per task." Because a more capable model like Astra may complete a task in fewer steps with less need for human correction or retries, it could prove more cost-effective than a cheaper model that requires extensive oversight.
As Astra begins its rollout, the ultimate test will be how quickly organizations feel comfortable integrating these agents into their core operations. The transition to the AGI era, as envisioned by OpenAI, may not be defined by a single "Eureka" moment, but by a gradual, cumulative shift in how professional work is performed. For CIOs and business leaders, the coming year will likely be defined by a new set of questions: how much agency should an AI worker be granted, and how can they maintain oversight as those systems become increasingly autonomous? According to Brockman, we are entering a period where the answer to those questions will become the defining factor of organizational success.
