OpenAI Releases GPT-6 Astra, Claims AGI Era

OpenAI began rolling out Astra, its new GPT-6 model, on September 3, and said the release marks the arrival of what it calls the artificial general intelligence era.

The rollout is staggered and comes wrapped in cybersecurity restrictions the company says it has never applied before. OpenAI says Astra is the first model to reach the "Critical" cybersecurity capability threshold under its own Preparedness Framework — the tier reserved for models that open new pathways to severe harm rather than simply amplifying existing ones.

Companies enrolled in OpenAI's application-based cybersecurity program get access first, according to CNBC. The Financial Times reported OpenAI is claiming the model overtakes rival Anthropic and could be considered artificial general intelligence.

What Astra Is Supposed to Do

OpenAI told TechCrunch that Astra represents "a new frontier on computer and browser use," and that it handles tasks with unmatched "speed, accuracy, and safety." The Verge reported the company is calling it its "most intelligent" model yet. NBC News reported OpenAI says the model follows user intent more closely and crosses a critical frontier for cybersecurity capability.

That last part is the reason for the guardrails. By OpenAI's account, Astra can find and exploit previously unknown security flaws without step-by-step human direction. In internal testing it discovered and exploited two zero-day vulnerabilities and scored a perfect mark on ExploitBench, a 20-vulnerability hacking evaluation. OpenAI says Astra refused 91.5% of harmful cyber requests, against 59% for its earlier GPT-5.6 Sol model.

Access to the model's most advanced cybersecurity features is being limited to vetted infrastructure defenders, including the U.S. government.