GPT-6 Astra Launches as OpenAI’s Most Advanced AI Model Yet — But Its Rollout Starts With Controversy
OpenAI has officially unveiled GPT-6 Astra, a new generation of artificial intelligence that the company describes as its most intelligent and capable model to date. Announced on September 3, 2026, Astra is designed to go far beyond conventional chatbot interactions, with major improvements in computer use, coding, scientific research, professional work and cybersecurity.
The launch represents a significant shift in how OpenAI sees AI assistants. Rather than simply responding to questions, GPT-6 Astra is designed to work through complicated, multi-step tasks, interact with software and computers, conduct research, analyze information and help users complete practical work.
OpenAI has even positioned the release as potentially marking the beginning of a new era in artificial intelligence. However, the ambitious launch was quickly overshadowed by a frustrating rollout that left many paying ChatGPT users waiting for access.
What Is GPT-6 Astra?
GPT-6 Astra is OpenAI’s latest frontier AI model, built to handle complex tasks that require reasoning, planning and extended interaction with digital tools.
According to OpenAI, Astra delivers substantial improvements over GPT-5.6 Sol in areas including software engineering, computer use, mathematics, science, cybersecurity and professional workflows. The company says the model is particularly effective when a task requires multiple steps rather than a single response.
Astra can also maintain better awareness of an evolving task. This means that when users add requirements or change direction during a lengthy workflow, the model is designed to incorporate those changes without losing track of the original objective.
That capability could make AI substantially more useful for real-world professional work.
From Answering Questions to Doing the Work
One of the biggest changes with Astra is its emphasis on computer use.
Instead of merely explaining how a task should be completed, the model can interact with computer environments and help perform actions. This opens the door to workflows such as filling out forms, working with business software, navigating websites, conducting research and handling repetitive digital tasks.
OpenAI’s published evaluations show Astra outperforming GPT-5.6 Sol on several computer-use benchmarks. On OSWorld 2.0, for example, Astra recorded a 72.6% score compared with 65.7% for GPT-5.6 Sol. It also achieved 92.7% on ScreenSpot-Pro, compared with 76.9% for the previous model.
These capabilities point toward a future in which AI systems act less like traditional chatbots and more like digital employees capable of completing portions of an entire workflow.
A Major Step Forward for Coding
Software development is another area where OpenAI is placing considerable emphasis on Astra.
OpenAI calls GPT-6 Astra its strongest software-engineering model so far. The model is designed to write, test, debug and modify software while working through complex repositories and development environments.
On Terminal-Bench 4.0, Astra achieved 57.9%, compared with 37.3% for GPT-5.6 Sol. On DeepSWE v1.1, Astra scored 74.1%, also ahead of the previous model’s 72.7%.
Astra also introduces a new approach to handling long coding sessions in Codex. Rather than relying entirely on repeatedly compressing earlier context into summaries, the system can preserve notes and retrieve information from previous context windows. This can help developers maintain continuity during large refactoring projects, debugging sessions and other lengthy engineering tasks.
For programmers, that could mean fewer repeated explanations and less time spent getting an AI coding assistant back up to speed.
Astra Pushes AI Into Scientific Research
OpenAI is also presenting GPT-6 Astra as a tool for scientific discovery.
The company reports strong performance across mathematics, science and health-related evaluations. Astra achieved 97.6% on FrontierMath Tier 4 and 96.0% on GPQA Diamond in OpenAI’s published comparisons.
More importantly, the model is designed to combine reasoning with computer interaction. Researchers could potentially use AI to examine datasets, interact with specialized scientific software and explore experimental results rather than relying solely on text-based analysis.
OpenAI has also highlighted two mathematical results involving prime-number gaps. According to the company, Astra helped researchers establish a stronger result concerning small gaps between primes and improve a bound involving unusually large prime gaps.
These examples illustrate the direction OpenAI wants Astra to take: AI that does not merely summarize scientific knowledge but actively assists with parts of the research process.
Cybersecurity Is One of Astra’s Most Powerful — and Concerning — Capabilities
Cybersecurity is perhaps the most consequential area of GPT-6 Astra’s capabilities.
OpenAI says Astra has reached its “Critical” cybersecurity capability threshold. The model demonstrated the ability to identify vulnerabilities and develop functional exploits in hardened systems.
In OpenAI’s published evaluations, Astra scored 100% on ExploitBench, compared with 78.5% for GPT-5.6 Sol. It also recorded substantially higher results on several other cybersecurity tests.
NDTV reported that Astra discovered two previously unknown vulnerabilities during OpenAI’s internal evaluations.
Such capabilities could provide enormous benefits to defenders. An AI that can discover weaknesses before criminals do could help organizations identify security problems, test their systems and respond to vulnerabilities more quickly.
But the same capabilities create obvious risks. A highly capable model that can discover and exploit vulnerabilities could also become dangerous if its access, tools or safeguards are poorly controlled.
OpenAI says Astra was therefore developed with additional safety measures and extensive testing. Its published safety evaluations show significantly lower rates on several internal computer-use and circumvention tests than GPT-5.6 Sol.
Is GPT-6 Astra Actually AGI?
The Astra launch has reignited one of the biggest debates in artificial intelligence: when does an advanced AI system become artificial general intelligence?
OpenAI President Greg Brockman suggested that Astra could eventually be viewed as a milestone in the arrival of AGI. He described the model as a generational leap and said that people looking back years from now might identify Astra as the point at which AGI was created.
However, OpenAI’s official documentation does not formally label Astra an AGI system.
That distinction matters. AGI does not have one universally accepted technical definition, so describing a model as the beginning of the “AGI era” is partly a statement about how OpenAI interprets the significance of its capabilities.
What is clearer is that Astra represents a move toward increasingly autonomous AI systems capable of handling complicated tasks across multiple domains.
The Launch Did Not Go as Planned
Despite the impressive technical announcement, OpenAI’s release quickly encountered criticism.
The initial rollout was limited, with access beginning for certain enterprise customers, including organizations using OpenAI’s Daybreak cybersecurity platform. Many paying ChatGPT subscribers expected broader access, particularly Pro users, but were left waiting.
The uncertainty surrounding when subscribers would receive access led to frustration across the OpenAI community.
Sam Altman subsequently apologized for what he described as a “messy rollout.” The company acknowledged the problems and indicated that broader access would follow, with Pro subscribers expected to receive priority as the rollout expanded.
OpenAI also announced compensation in the form of banked usage resets for affected paid users. Users who remain without Astra access are expected to accumulate an additional reset for each day they are unable to use the model.
The situation highlights an increasingly important problem for AI companies: launching a frontier model is no longer simply about publishing a model. Companies must simultaneously manage enormous demand, infrastructure, safety requirements, user expectations and access tiers.
How Much Does GPT-6 Astra Cost?
For developers, OpenAI has announced API pricing of $10 per million input tokens and $50 per million output tokens under its standard processing mode.
A faster processing option is also available at twice the standard price and can provide up to twice the speed.
Astra is being made available through OpenAI’s API and is also planned for deployment through Microsoft Azure and Amazon Bedrock.
The pricing puts Astra firmly in the premium frontier-model category, reflecting the computational resources required to operate a model designed for complex reasoning and long-running agentic workloads.
What GPT-6 Astra Means for the Future of AI
The significance of GPT-6 Astra goes beyond benchmark scores.
Previous generations of AI largely revolved around generating information: answering questions, writing text, producing images or creating code.
Astra points toward a different model of AI — systems that can understand an objective, plan a sequence of actions, use digital tools and work toward completing the objective.
That shift could transform numerous industries.
Businesses could use AI agents for research, data analysis and administrative work. Developers could delegate increasingly complex programming tasks. Scientists could use AI to explore datasets and research problems. Cybersecurity teams could employ AI to find vulnerabilities before attackers discover them.
At the same time, increasingly autonomous AI creates new challenges involving security, reliability, oversight and accountability.
A Powerful Launch With an Important Warning
GPT-6 Astra may ultimately be remembered as one of the major milestones in OpenAI’s development of increasingly autonomous AI.
Its improvements in coding, computer use, scientific reasoning and cybersecurity suggest that frontier models are moving beyond the role of conversational assistants. They are becoming systems capable of participating directly in complex workflows.
But the messy launch also provides a reminder that technological capability and reliable deployment are two different challenges.
OpenAI now has to prove that it can make Astra not only powerful, but also accessible, dependable and safe at scale.
For users and businesses watching the AI race, GPT-6 Astra represents both sides of the next stage of artificial intelligence: unprecedented capability paired with unprecedented responsibility.