The long-circulating rumors have finally been confirmed, and the implications for the technology sector are profound. OpenAI has officially released GPT-6 Astra, a sophisticated new frontier model that the company suggests represents the onset of artificial general intelligence (AGI). For years, OpenAI has maintained a clear, ambitious charter: to develop highly autonomous systems capable of outperforming humans at the vast majority of economically valuable work. With the arrival of Astra, the company believes it has reached a critical, long-sought milestone in that mission.
During a closed-door press briefing held earlier today, OpenAI co-founder and president Greg Brockman provided a remarkably direct assessment of the launch. He concluded the session with a statement that serves as a landmark for the industry: “Welcome to the AGI era.” Such a bold framing is rare, even by the standards of the hyper-competitive AI landscape, where superlative claims are common. However, for enterprise leaders, the significance of Astra is less about philosophical definitions of intelligence and more about a shift in the fundamental nature of computing.
OpenAI is positioning GPT-6 Astra as a departure from the keyboard-and-mouse paradigm that has dominated personal computing for decades. The company’s internal launch materials describe the technology as “the world’s best computer use model.” Rather than requiring developers to laboriously build and maintain dedicated API integrations for every piece of software a company uses, Astra is engineered to navigate software interfaces much like a human does. It can operate across web browsers, spreadsheets, desktop applications, and specialized engineering software, carrying out multistep workflows and producing finished documents or presentations rather than simply offering instructions on how a human might complete them.
To illustrate this shift, OpenAI released a promotional video that juxtaposes a 1980s AI demonstration—in which a computer is asked to draw a simple yellow circle—with the capabilities of Astra. In the modern demo, employees use voice commands to instruct Astra to transform a yellow circle into a complex rocket ship, evolve it into a full 3D game in a matter of minutes, and even manage an eBay listing. The demonstration highlights a transition from a model that answers questions to one that executes intent.
Astra is scheduled to begin its rollout this Thursday to enterprise customers through OpenAI’s gated access program, known as Daybreak. The company plans to expand availability over the coming days to include ChatGPT Plus, Pro, Business, and Enterprise users, as well as providing access via the OpenAI API and cloud platforms including Amazon Web Services (AWS) Bedrock and Microsoft Azure.
From Answering Questions to Operating Computers
The primary value proposition for the enterprise lies in this newfound ability to perform computer work. OpenAI claims the model can handle a vast array of administrative and technical tasks, including filling out online forms, updating CRM records, managing calendars, conducting complex web research, and drafting results into documents or emails. Furthermore, the model can manipulate spreadsheets, analyze scientific data in Python notebooks, operate in Power BI, build and test websites, and troubleshoot software or engineering applications like KiCad and FreeCAD.
This represents a potential pivot in how enterprise AI architecture is designed. For much of the recent generative AI boom, corporations have been forced to bridge the gap between models and their proprietary systems through complex webs of APIs, plugins, and retrieval systems. Brockman argued that computer-use agents could bypass much of this friction because software is already designed to be navigated by a human. By leveraging that existing interface, the agent can, as Brockman put it, “zip through spreadsheets, fill out forms, and navigate across web pages.”
OpenAI’s internal testing suggests significant performance gains. On an offline subset of OSWorld 2.0, Astra achieved a score of 72.6% while requiring approximately 40 minutes per task. In comparison, the previous model, GPT-5.6 Sol, scored 65.7% but required roughly 75 minutes, marking a 47% reduction in time per task. Beyond mere speed, the company demonstrated the model’s ability to multitask—creating a 3D game while simultaneously preparing a legal agreement—marking a move away from the chatbot pattern where humans must provide constant, iterative instruction. OpenAI researcher Mia Glaese noted that this shift encourages a new management style: humans will now direct work at a higher level, delegating complex multi-application workflows to the AI.
The Scale of Training and the AGI Debate
Aidan Clark, a researcher at OpenAI, described the development of Astra as the company’s largest-scale training run to date. It is the first OpenAI model pretrained using more than 100,000 DBUs at the company’s Stargate infrastructure. Notably, it is also the first model for which previous iterations played a supervisory role in the training process. According to Clark, the jump in capability from the previous model to Astra is significantly larger than the improvements seen in prior generations.
The benchmark results are impressive, with Astra scoring 97.6% on FrontierMath Tier 4 v2, 74.1% on DeepSWE v1.1, 95.9% on BenchCAD, 96% on GPQA Diamond, and 100% on ExploitBench. Perhaps most notably, it reported a 98.6% score on ARC-AGI-3. However, this figure has ignited debate within the AI community regarding how intelligence is measured.
The ARC-AGI-3 benchmark is designed to test an AI’s ability to generalize to unfamiliar problems. While Astra’s score is high, comparisons are difficult because the surrounding system architecture—such as persistent memory, tool-use, and feedback mechanisms—can dramatically influence results. For example, NVIDIA recently reported that its AVO architecture reached a 100% score on the same benchmark, though it relied on an underlying model with a base capability of roughly 30%. This has led to an ongoing debate about whether the "intelligence" resides in the foundation model or the entire agentic system.
For businesses, this distinction may be secondary to results. As Brockman noted, the path to AGI has not been a single, well-defined moment, but rather a "gray, fuzzy thing." While he personally believes the arrival of Astra signals that we have reached the AGI era, he acknowledges that the practical measure of such a claim will be found in the reliability and utility of the work these systems can perform.
Economics and Governance in the Age of Agents
OpenAI is also shifting the narrative around the cost of AI. While traditional pricing is based on tokens, Brockman argues that this is an increasingly poor metric for enterprise value. Instead, businesses should evaluate the "price per task." Because a more capable model like Astra can complete a workflow more efficiently and with fewer errors than a cheaper, less capable model, the total cost of ownership may actually be lower.
However, this increased capability introduces a significant governance challenge. Unlike a chatbot that generates text for human review, an agent operating a computer can actively manipulate files, send information, and change records. To address this, OpenAI is implementing a "defense-in-depth" safety strategy. This includes internal alignment, security classifiers, and monitoring systems designed to ensure the model stays within its authorized boundaries.
During a period of internal security hardening, OpenAI briefly paused some training runs to ensure its safety and monitoring infrastructure kept pace with model capabilities. This reflects an enterprise-level approach to risk management. The goal is to train the agent to recognize when a task requires it to exceed its authorization and, rather than attempting to bypass security controls, return to the human user for further direction.
As OpenAI continues to navigate these complexities, the company’s leadership remains committed to the principle that intelligence and alignment must advance in tandem. Jakub Pachocki, OpenAI’s chief scientist, emphasized that progress in intelligence does not guarantee safety. Consequently, the company has pledged to slow or halt scaling if its confidence in model alignment falls below acceptable levels.
As Astra begins its deployment, the focus will likely shift from academic benchmarks to the real-world utility of autonomous agents. For organizations, the era of AGI may not be marked by a single event, but by a gradual transition where AI becomes a reliable, albeit highly monitored, worker capable of executing complex objectives. Whether Astra is officially labeled AGI or not, its ability to perform high-level professional work represents a turning point in the digital economy.
