4 hours ago
14
OpenAI’s GPT-6 Astra is live. The company’s most powerful AI model started reaching users on September 3, beginning with a small group of testers before expanding to paid ChatGPT subscribers across Plus, Pro, Business, and Enterprise tiers.
Multiple users confirmed the rollout in real time, reporting that Astra appeared in their accounts within minutes of the initial deployment. The launch marks what OpenAI president Greg Brockman called a “generational leap” and the beginning of what the company considers the AGI era.
What makes Astra different
The performance numbers tell the story pretty clearly. Astra completes tasks approximately 47% faster than GPT-5.6 Sol, its immediate predecessor. On the OSWorld 2.0 benchmark, Astra scored 72.6% in 40 minutes, compared to Sol’s 65.7% in 75 minutes.
Astra scored 98% on FrontierMath Tier 4 and a perfect 100% on ExploitBench. That last number means the model can autonomously detect and exploit unknown vulnerabilities in secure systems.
That capability earned Astra OpenAI’s highest internal classification: “Critical” on its Preparedness Framework for cybersecurity. It’s the first model to hit that threshold. OpenAI’s response has been to restrict access to the most advanced cybersecurity features during this initial phase, with plans to eventually broaden availability for defensive applications.
A rocky first 24 hours
OpenAI initially prioritized access for organizations enrolled in its Daybreak cybersecurity program, along with enterprise customers. By September 4, enough ChatGPT Plus and Pro users had reported access issues that CEO Sam Altman posted a public apology. He acknowledged the confusion, particularly among subscribers who expected immediate access to a model included in their existing subscription plans. OpenAI offered banked resets as compensation for the service interruptions.
Astra is included in existing subscription tiers at no additional cost, so the bottleneck was rollout logistics, not pricing or licensing.
The cybersecurity paradox
Astra’s perfect ExploitBench score puts OpenAI in a position that few AI companies have had to navigate at this scale. The Preparedness Framework’s “Critical” designation triggers additional safety protocols and access controls before features reach a broad audience.
What this means for the AI landscape
Companies already using ChatGPT for business operations now have access to a significantly more capable system at no additional cost. Brockman’s framing of Astra as the start of the AGI era is ambitious branding, but the benchmarks give the claim more substance than most such declarations tend to carry.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our Editorial Policy.
English (US) ·