OpenAI Declares 'AGI Era' With GPT-6 Astra, Its Most Powerful and Most Tightly Guarded Model Yet
OpenAI Declares ‘AGI Era’ With GPT-6 Astra, Its Most Powerful and Most Tightly Guarded Model Yet #
OpenAI launched GPT-6 Astra on Thursday, its most capable AI model to date, with company president Greg Brockman declaring the dawn of an “AGI era” and describing the release as a “generational leap” for artificial intelligence. It is the company’s most direct claim yet that it is approaching, or has crossed, the threshold of artificial general intelligence, broadly defined as systems that can outperform humans at most economically valuable work.
Rollout began Thursday with enterprise customers enrolled in OpenAI’s Daybreak Access program, an application-based tier for approved business testers. Broader availability for ChatGPT Plus, Pro, Business and Enterprise subscribers, as well as through the OpenAI API, Microsoft Azure and AWS Bedrock, is expected within days, according to the company’s launch announcement.
In a pre-launch media briefing, Brockman said he personally believes Astra may mark the arrival of AGI, closing the session with the declaration: “Welcome to the AGI era.” CEO Sam Altman, speaking in a separate interview, said the extended development timeline would prove justified: “We think it’ll be worth the wait.”
According to OpenAI’s technical documentation, Astra was trained on the company’s largest-ever compute run, drawing on more than 100,000 GPUs at its Stargate facility in Texas. It is also the first model for which OpenAI used other AI systems in a significant supervisory capacity during training, a structural change the company says drove both capability and alignment gains.
On published benchmarks, Astra scores 99.9% on ARC-AGI-3, 98% on FrontierMath Tier 4 and a perfect 100% on ExploitBench. OpenAI calls it the world’s best computer use model, capable of navigating browsers, desktop applications and spreadsheets to produce finished documents and execute multistep professional workflows autonomously.
The most contentious aspect of the release is Astra’s cybersecurity profile. OpenAI has designated Astra as the first model to reach a “critical” tier under its internal preparedness framework, meaning it can independently identify previously unknown software vulnerabilities and build novel exploit chains in hardened systems without step-by-step human direction. The company said it discovered this capability during testing and deliberately delayed the launch to conduct additional safety work.
To contain the risk, Astra’s most advanced offensive cybersecurity functions will remain restricted to a small set of vetted organizations in OpenAI’s application-based cybersecurity access program. Even for general enterprise and API users, additional automated safety checks may pause or halt certain sensitive tasks, the company said.
The release puts OpenAI in direct competition with Anthropic and other frontier AI labs. The company asserts that Astra sets a new standard across professional domains including software engineering, financial modeling and scientific research.
Copy link