Edgar Cervantes / Android Authority
Add Android Authority on Google:
TL;DR
- OpenAI has launched GPT-6 Astra, with stronger capabilities in software engineering, computer use, and cybersecurity.
- Astra is OpenAI’s first model to reach its “Critical” cybersecurity threshold and is more jailbreak-resistant than GPT-5.6 Sol.
- OpenAI isn’t officially calling Astra AGI, but its president says he personally thinks “we’re there.”
OpenAI has launched GPT-6 Astra, its latest flagship AI model and what the company describes as the most capable model it has deployed to date. Astra brings notable gains in areas such as software engineering, computer use, and cybersecurity, and it will roll out to paid ChatGPT users and the API over the coming week. And while OpenAI isn’t officially calling this AGI, one of its founders now says he personally thinks we’ve reached that point.
As reported by TechCrunch, OpenAI is particularly talking up Astra’s ability to use computers and browsers, while president Greg Brockman described it as the company’s most intelligent model yet. The company also reckons Astra is its best model so far for software engineering, with testing showing improvements in areas like bug detection and working with codebases.
A factor that many won’t find reassuring is that Astra is the first OpenAI model to hit the “Critical” level for cybersecurity capabilities under the company’s Preparedness Framework. OpenAI says that when given the right tools and access, Astra can find previously unknown security flaws and develop ways to exploit well-protected systems without requiring a human to guide every step. That’s potentially very useful for finding vulnerabilities before attackers do, but the downside probably doesn’t need much explaining.
OpenAI says Astra is also more resistant to jailbreak attempts than GPT-5.6 Sol, while its broader safety testing generally shows fewer signs of problematic behavior. Given Astra’s stronger capabilities, the company is also applying additional monitoring to tool-based sessions as an extra layer of protection.
The less-than-comforting developments continue when it comes to actually keeping an eye on what Astra is doing. TechCrunch highlights the model’s use of a reasoning technique called opaque recurrence, which can make its chain of thought harder for researchers to follow. OpenAI chief scientist Jakub Pachocki acknowledged that monitoring is becoming more difficult as models improve, partly because more capable models can complete harder tasks using fewer language tokens, or sometimes none at all. OpenAI’s system card supports that broader concern, finding that Astra’s monitorability has decreased relative to Sol.
Which brings us back to AGI. When asked during a media call whether Astra represented its arrival, Brockman said AGI is no longer a contractual trigger for OpenAI and has become more of a “mission concept or spiritual concept.” He stopped short of declaring Astra to be AGI on OpenAI’s behalf, but his personal view was considerably less ambiguous: “For me personally, I do think we’re there.”
Thank you for being part of our community. Read our Comment Policy before posting.