On September 3, 2026, OpenAI released GPT-6 Astra — a model the company calls a “generational leap” and the most powerful development in its entire history. We break down what the new model can do and why its release was accompanied by unusual caution from the company itself.
What Is GPT-6 Astra
Astra is OpenAI’s new flagship, which the company describes as “the smartest and best-aligned model in the world.” According to the company, Astra demonstrates the highest level among all models in working with computers, browsers, software development, cybersecurity, science and professional tasks.
OpenAI president Greg Brockman said that the model may eventually be seen as the moment when AGI (artificial general intelligence) emerged. According to CEO Sam Altman, Astra has taken him to a “new level of capabilities” and has already changed his own workflows.
A Record in Mathematics and New Capabilities
Astra achieved 98% on FrontierMath Tier 4 — one of the most difficult mathematics benchmarks, developed specifically to test problems that cannot be solved through memorized solutions. According to OpenAI, the model has already helped solve long-standing open problems in mathematics.
A separate new feature concerns Codex: instead of the standard “compression” of conversation history when the context window fills up (when the model simply summarizes previous work), Astra received a new way to preserve and restore context during long sessions — for example, during complex debugging or large-scale code refactoring.
Why the Release Was Delayed and What “Critical” Cyber Risk Means
The most important nuance of this release is not the capabilities, but the caution with which OpenAI approached it. The company initially planned a public release for September 5, but delayed it after an incident involving a leak on Hugging Face in July 2026 in order to add additional safeguards.
On September 1, OpenAI officially reported that, based on additional testing, Astra meets the threshold for a “Critical” level of cyber capability under the company’s internal Preparedness Framework — the first OpenAI model to receive this classification. In simple terms, with the necessary tools and access, the model could theoretically independently, without step-by-step human guidance, find previously unknown vulnerabilities and develop ways to exploit them in well-protected systems.
Because of this, access to the model’s most advanced cybersecurity capabilities will be significantly restricted. For now, Astra is available only to customers of the Daybreak cybersecurity program — companies that undergo a separate verification process.
When Astra Will Become Available to Regular Users
In the coming days after the initial release, Astra will become available to users of ChatGPT Plus, Pro, Business and Enterprise plans, as well as through the OpenAI API and Amazon Web Services.
In Brief: The Key Points
- September 3, 2026 — GPT-6 Astra release, initially with limited access through the Daybreak program
- In the coming days — access for Plus, Pro, Business, Enterprise and through the API
- 98% on FrontierMath Tier 4 — a record result in mathematics
- A new way to preserve context in Codex for long coding sessions
- The release was delayed because of the July incident on Hugging Face — safeguards were added
- Astra is the first OpenAI model to receive “Critical” cyber-risk status under the Preparedness Framework; access to advanced capabilities is restricted
FAQ
What is GPT-6 Astra? OpenAI’s new flagship AI, released on September 3, 2026 — according to the company, the most powerful model in OpenAI’s history, setting records in mathematics, computer use and software development.
When will GPT-6 Astra become available to regular users? Within a few days after the September 3 release — for ChatGPT Plus, Pro, Business and Enterprise plans, as well as through the API and AWS.
What does “Critical” cyber-risk status mean? This is the highest category under OpenAI’s internal risk assessment system (Preparedness Framework). Astra is the first model from the company to reach this level, so access to the model’s most advanced capabilities is significantly restricted and available only to verified organizations.
Why was the Astra release delayed? Because of an incident involving a leak on Hugging Face in July 2026 — the company decided to add additional safeguards before the public release.
How well does Astra perform in mathematics? The model scored 98% on FrontierMath Tier 4 — one of the most difficult mathematics benchmarks, and according to OpenAI, it has already helped solve long-standing open mathematical problems.
The article was prepared by the TechVisor team — practical IT media for people.
