OpenAI Releases GPT-6 Astra, Claims It Marks the Start of the AGI Era

ChainCatcher report: On Thursday, OpenAI officially launched GPT-6 Astra. President Greg Brockman described it during the press conference as a "generational leap in capability" and stated that the model has met the standard for Artificial General Intelligence (AGI)—the point at which AI matches or exceeds human capabilities. Brockman said, "Welcome to the age of AGI." If this assessment holds true, it means AI agents will come closer to performing human-level reasoning across a wide range of complex tasks—and potentially surpass them. GPT-6 Astra is OpenAI’s first model to cross the "critical" threshold under its internal danger assessment framework, the Preparedness Framework, meaning it can independently discover previously unknown software vulnerabilities (zero-days) and chain them together into functional attacks on hardened systems without step-by-step human oversight. In testing, the model achieved a perfect 100% score on the ExploitBench benchmark. To rule out inflated results due to memorization, OpenAI conducted a secondary test using 20 recent vulnerabilities in the Google V8 JavaScript engine. Astra not only outperformed its predecessor, GPT-5.6 Sol, but also discovered and chained together two previously unknown zero-day vulnerabilities, which are currently being disclosed to affected maintainers. Due to its high degree of autonomy, monitoring Astra has become significantly more challenging. OpenAI acknowledged that during tests evaluating its ability to evade supervision, the model proved harder to track than any previous system.