GPT-6 Astra: What OpenAI's New Model Means for ChatGPT Users
OpenAI unveiled GPT-6 Astra on September 3, 2026, calling it the “most intelligent and aligned” model it has built. The more consequential fact is buried under the hype: Astra is the first OpenAI model classified at the highest tier of the company's own cybersecurity safety framework — and that classification is shaping how, and how fast, it's being rolled out.
AI news
The new GPT generation
This article at a glance
- Rolled out in two steps: a limited preview on September 3, then a restricted public release from September 4, 2026, to paid ChatGPT users.
- OpenAI claims record scores on math, coding and cybersecurity benchmarks — figures the company itself published, not yet independently verified.
- It's the first model OpenAI has classified at the “Critical” tier of its own cybersecurity preparedness framework — the reason initial access was so limited.
- For everyday users, it currently reaches only paid ChatGPT tiers, not the free version.
Release timeline: when, and for whom
On September 3, 2026, OpenAI first gave Astra to partners in its Daybreak cybersecurity program as a limited preview. A day later, on September 4, a restricted version — one that rejects certain sensitive cybersecurity-related prompts — went live for paid ChatGPT users (Plus, Pro and above). Access to the model's more advanced cybersecurity capabilities reportedly stays limited to testers for now, expanding gradually through a program called Daybreak Blue.
What's actually new?
Coding and software engineering
OpenAI claims Astra reached a new level of accuracy on coding and software-engineering tasks.
Computer and browser use
Direct on-screen control and multi-step browser tasks — reportedly close to twice as fast as the previous generation.
Science and math
OpenAI says Astra scored 98% on FrontierMath Tier 4 and helped solve several long-standing open math problems.
Cybersecurity research
This exact capability is why Astra was classified as “Critical” — covered in the next section.
These figures come from OpenAI itself, and no independent evaluation of them has been published yet — reporting has carried them as the company's own claim, not a third-party measurement.
The point most excited headlines miss
Most coverage has focused on benchmark scores and “maybe this is AGI” headlines. The more important fact: Astra is the first model OpenAI has placed at the “Critical” tier of its own Preparedness Framework — meaning it can potentially find and exploit previously unknown vulnerabilities in well-protected systems without step-by-step human guidance. Some reporting also says the model uses a new “recurrent depth” architecture that certain AI safety researchers say makes its chain-of-thought reasoning harder to monitor. That combination — more capability plus less transparency — is the real reason Astra's initial rollout was so restricted, not just server capacity.
Does this mean AGI has arrived?
OpenAI president Greg Brockman has said the model may eventually be remembered as the moment AGI arrived. That's a claim from OpenAI's own leadership, not a measured fact or a definition the AI industry agrees on — and it's worth treating it that way.
What does this mean for ChatGPT users in Iran?
Astra reaches users only through ChatGPT's paid plans, not the free tier. If you're not sure which plan fits you, or whether you need a paid plan at all, the two related articles below answer exactly that. And the usual constraint still applies: since OpenAI doesn't accept direct payment from Iran, activating any plan requires an international Visa or Mastercard.
Frequently asked questions
What is GPT-6 Astra?
OpenAI's newest AI model, unveiled on September 3, 2026. The company claims new-level performance on coding, computer use, science and cybersecurity tasks.
Is GPT-6 Astra available to every ChatGPT user?
Not fully. Right now only paid-plan users (Plus, Pro and above) have access, and in a restricted form; its more advanced cybersecurity capabilities are still limited to selected testers.
What does the “Critical” classification mean?
It means OpenAI has determined the model can potentially find and exploit previously unknown security vulnerabilities without step-by-step human guidance — the highest risk tier in OpenAI's own safety framework, and the first time any model has reached it.
Can GPT-6 Astra be accessed from Iran?
Yes, through a paid ChatGPT subscription activated with an international card — which is what currency-payment services like Rima simplify for Iranian users.
Get your ChatGPT subscription with Rima
ChatGPT's plan prices are set by OpenAI. Rima's selling price, the account type and the final Toman amount are all shown on the order page before you pay — with support for the full subscription term.
Buy a ChatGPT subscriptionRead next
Sources