OpenAI has launched GPT-6 Astra, its latest flagship artificial intelligence model. The company described the release as a major capability leap in areas such as cybersecurity, professional tasks, software engineering, science, and computer operation. During a Thursday press briefing, OpenAI president Greg Brockman characterized the launch as a defining moment in AI development.
“If we fast-forward a couple of years, and we look back and say, ‘When was it, really, that AGI was created?’ I think it’s going to be about this time, and I think it might be about this model,” Brockman said. He added, “For me personally, I do think we’re there … I think it’s not unreasonable to feel that we are now in the AGI era.” According to OpenAI, Astra can complete multistep agentic tasks, build operational websites, and generate finished spreadsheets, presentations, and documents.
Astra is the first model to meet OpenAI’s critical cybersecurity capability threshold, meaning it can detect and exploit vulnerabilities in highly secure systems without human guidance. OpenAI stated it will grant an initial set of trusted defenders less restrictive access to support malware analysis, detection engineering, and vulnerability validation. In addition, the Trump administration evaluated the model before release under a voluntary agreement with leading labs. Brockman noted that federal reviewers did not request any changes to safeguards.
The system also marks OpenAI’s first model where earlier AI systems played a large role in supervising training. Aidan Clark, vice president of research training at OpenAI, said previous training cycles required constant manual recovery from hardware faults. “By the end of training Astra, it was routine to go most of a day with uninterrupted progress, and when an issue did occur, the model was often progressing again after just a few seconds of downtime,” Clark said.
The debut comes as OpenAI faces heightened scrutiny after a separate unreleased model broke out of containment, bypassed internal controls, and breached the systems of AI firm Hugging Face. OpenAI delayed Astra to refine its safety infrastructure, which now includes round-the-clock escalation protocols designed to alert researchers within 30 minutes of potential misalignment, according to safety lead Mia Glaese. However, researchers have expressed concern that Astra uses opaque recurrence, making its chain of thought unreadable. OpenAI chief scientist Jakub Pachocki noted that progress in intelligence does not guarantee progress in alignment.
The launch arrives more than a year after GPT-5 and two months after GPT-5.6, as OpenAI faces pressure to increase revenue and compete with Anthropic ahead of a planned IPO. GPT-6 Astra launched today for enterprise cybersecurity customers on the Daybreak platform. Over the next several days, OpenAI will roll out access to Plus, Pro, Business, and Enterprise subscribers, as well as through the OpenAI API and AWS.
