The rumors that have swirled through the tech industry for months have finally been confirmed: OpenAI has officially unveiled GPT-6 Astra. Described by the company as its most significant frontier model to date, Astra is being presented as the technical realization of Artificial General Intelligence (AGI)—the long-sought, highly autonomous system capable of outperforming humans at most economically valuable work. During a closed-door press briefing held earlier today, OpenAI co-founder and president Greg Brockman provided a definitive, if understated, assessment of the milestone. He ended the session with a simple, resonant statement: “Welcome to the AGI era.”
This framing represents a dramatic shift in how the industry discusses generative AI. While previous launches focused on the ability of models to answer questions or generate text, OpenAI is positioning GPT-6 Astra as the first true “computer use” model. The core promise of the technology is that users—ranging from individual employees to large-scale enterprises—will no longer need to rely on the traditional, manual cycle of clicking a mouse and typing on a keyboard to interface with software. Instead, Astra is designed to navigate complex digital environments just as a human would, moving seamlessly across browsers, spreadsheets, professional applications, and operating systems to execute multi-step workflows.
In promotional materials released alongside the announcement, OpenAI dubbed the new model “the world’s best computer use model.” The company demonstrated this capability with a video that contrasted an early 1980s AI demo—which could perform the rudimentary task of drawing a yellow circle—with the modern reality of Astra. In the demonstration, OpenAI employees used voice commands to direct the model to transform that same yellow circle into a complex rocket ship design, build a functional 3D game in a matter of minutes, and even navigate to eBay to create a product listing, all without the human user ever touching a keyboard.
For enterprise customers, the rollout begins this Thursday through OpenAI’s gated access program, Daybreak. The model will then be made available over the coming days to ChatGPT Plus, Pro, Business, and Enterprise subscribers, as well as through the OpenAI API and major cloud platforms including AWS Bedrock and Microsoft Azure.
From Chatbots to Digital Operators
The enterprise case for GPT-6 Astra is centered on its ability to perform actual work rather than merely providing information. OpenAI claims the model can autonomously fill out online forms, update CRM records, manage complex calendars, conduct exhaustive web research, and synthesize results into finished documents or emails. Beyond basic office tasks, the model is capable of manipulating spreadsheets, performing scientific data analysis in Python notebooks, operating professional engineering software like KiCad and FreeCAD, and even troubleshooting technical software installations.
This evolution points toward a fundamental change in enterprise AI architecture. Since the inception of the generative AI boom, businesses have been forced to build complex, brittle pipelines connecting models to corporate systems via APIs, plugins, and retrieval-augmented generation (RAG) systems. Brockman argued that computer-use agents like Astra could bypass much of this painstaking integration work. Because modern software is already designed with the human user in mind, an agent that can interact with the same interface as a human effectively turns the entire digital world into an 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 advent of computer-use capabilities, an agent can simply “zip through spreadsheets, fill out forms, and navigate across web pages” just like a skilled employee. According to OpenAI, this capability has been in the works since the company’s earliest days, when researchers first theorized about training an agent that perceives the same basic inputs and outputs as a human: pixels, keyboards, and mice.
The performance gains are stark. On an offline subset of the OSWorld 2.0 benchmark, Astra achieved a success rate of 72.6% while taking an average of 40 minutes per task. In comparison, the previous iteration, GPT-5.6 Sol, scored 65.7% while requiring roughly 75 minutes per task—a 47% reduction in time. Beyond speed, OpenAI showcased the model’s ability to handle complex, concurrent tasks, such as creating a 3D game while simultaneously drafting a legal agreement.
Scaling the Frontier
Aidan Clark, an OpenAI researcher, characterized the development of Astra as the company’s largest-scale training run to date. It is the first model trained using over 100,000 DBUs at the company’s Stargate infrastructure. Furthermore, it marks a milestone in self-improving systems, as previous models played a critical, supervisory role in the training of Astra. Clark noted that internal evaluations suggest the performance leap from Sol to Astra is significantly larger than the jump from previous generations to Sol.
The benchmark results for Astra are wide-ranging. It achieved 97.6% on FrontierMath Tier 4 v2, 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, a benchmark designed to measure an AI’s ability to generalize to unfamiliar problems rather than simply regurgitating training data.
However, these figures have ignited a debate within the AI community regarding how intelligence should be measured. Comparisons are complicated by the fact that Astra uses OpenAI’s specific “Responses API” harness, while other models may operate under different configurations. Recent results from NVIDIA, which used an agentic architecture called AVO to reach 100% on the ARC-AGI-3 benchmark using a model that had a 30% baseline, illustrate that performance is often a result of the entire system—including memory, tool use, and feedback loops—rather than the raw neural weights of the foundation model alone.
This disagreement highlights the "gray, fuzzy" nature of defining AGI, as Brockman put it. While the industry debates whether AGI is a specific threshold or a moving target, OpenAI is pivoting toward a more pragmatic definition. For the company, and for the enterprise customers they serve, the distinction between a "pure" foundation model and a "complete agent system" may matter less than the reliability, cost, and auditability of the work being produced.
The Governance Challenge
As Astra moves from a chatbot that offers advice to an agent that takes action, the risks involved in its deployment have grown substantially. An agent operating a computer has the potential to alter records, transmit sensitive data, or interact with critical systems. Mia Glaese, an OpenAI researcher, emphasized that as models become more autonomous, the company’s understanding of alignment and safety must evolve in tandem.
To manage these risks, OpenAI has implemented a “defense-in-depth” security strategy. During the development of Astra, the company intentionally paused some training operations for two weeks to tighten infrastructure controls and ensure that safety measures were keeping pace with the model’s rapidly advancing capabilities. These measures include advanced classifiers and offline detection systems designed to identify malicious patterns of behavior across multiple prompts—a necessity when dealing with autonomous agents that can perform long-horizon tasks.
The company is particularly focused on "observability," or the ability to monitor the reasoning of a model as it works. Jakub Pachocki, OpenAI’s chief scientist, warned that progress in intelligence does not inherently guarantee progress in alignment. He stressed that OpenAI is prepared to slow or halt scaling if the company cannot maintain sufficient confidence in its ability to monitor and govern the model’s behavior.
A Shift in Economic Value
Finally, the introduction of Astra represents a pivot in how OpenAI wants users to perceive the cost of AI. Brockman argued that token-based pricing is becoming an outdated metric for enterprise value. Instead, he suggested that the industry should focus on "price per task." By automating workflows that previously required significant human labor, Astra aims to reduce the overall cost of operations, even if the model itself requires more computational power per inference.
As businesses begin to integrate these agents into their core workflows, the ultimate measure of Astra’s success will be the degree to which they trust it with meaningful, independent work. Whether or not it is labeled as "AGI," Astra represents a shift in the way organizations interact with software. As Brockman noted, the arrival of this era may not be marked by a single, definitive moment, but rather as an economic transition that becomes clear only when viewed in retrospect—a world where human work is defined by high-level direction, and the execution is handled by an increasingly autonomous, intelligent system.
