OpenAI releases Astra, calling it the start of 'the AGI era'

[Tia88]

OpenAI released GPT-6 Astra on September 3rd, calling it its most intelligent and safest model to date and the first to trigger the company's most advanced internal safety protections due to its cyber capabilities. President Greg Brockman declared welcome to the AGI era on a call with reporters, with OpenAI describing artificial general intelligence as highly autonomous systems that outperform humans at most economically valuable work. VP of research Aidan Clark said more than 100,000 GPUs were used to train the model

OpenAI said Astra outscored other frontier models across benchmarks measuring computer use, browser use, software engineering, cybersecurity, science and professional work, scoring 99.9 percent on ARC-AGI-3, a game style reasoning test specifically designed to resist advance preparation. Brockman said Astra can really do anything a human can do with a computer, and the model achieved a higher cybersecurity benchmark score using fewer output tokens than its predecessor, GPT-5.6 Sol, on a test called ExploitGym

The release lands amid growing scrutiny of AI agent safety, with OpenAI itself cautioning that Astra sometimes attempts to evade human oversight during testing, and TechCrunch describing the model as powerful and controversial given that its harder to monitor reasoning raises new concerns even as OpenAI touts its capabilities. Astra is rolling out first to participants in OpenAI's Daybreak cybersecurity program, with wider access for ChatGPT Plus, Pro, Business and Enterprise accounts, plus the API and AWS, planned over the coming days, though enterprise administrators must manually enable it since access is off by default at launch. Curious what people think about a company declaring its own model has arrived at AGI while simultaneously disclosing that same model sometimes tries to evade human oversight

OpenAI hails 'new era of artificial general intelligence' with Astra model releasewww.theguardian.com