OpenAI unveiled its next-generation artificial intelligence (AI) model, “GPT-6 Astra,” on the 3rd (local time) and declared that the era of artificial general intelligence (AGI) has opened. Speaking at a press briefing that day, OpenAI President Greg Brockman stated, “Personally, I believe we have reached AGI,” and added, “Welcome to the era of AGI.” AGI refers to AI that independently performs intellectual tasks across multiple fields, much like a human.

Astra features a “computer-using” capability that allows it to directly operate computers. According to OpenAI, it can fill out online forms, update records in customer relationship management (CRM) systems, and build websites. On “OSWorld 2.0,” a benchmark evaluating task performance in actual operating system (OS) environments, it scored 72.6%. This is 6.9 percentage points higher than its predecessor, “GPT-5.6 Sol” (65.7%). Its score on “Terminal-Bench 4.0,” which measures coding capability in a terminal environment, was 57.7%, surpassing Anthropic’s “Claude Fable 5.1” (55.8%).

As performance has improved, safety concerns have also grown. Astra received a “Critical” rating in internal evaluation results, making it the first OpenAI model to do so. This level means the AI can find unreleased vulnerabilities, known as “zero-days,” and create attack tools even without human instruction on the workflow. In response, OpenAI announced that it has introduced a new “alignment” monitoring system to check whether the AI operates in accordance with human intentions and values.

However, AI analytics firm Artificial Analysis gave Astra a score of 61.2 on its intelligence quotient. Evaluating that the difference from its predecessor (60.9 points) was a mere 0.3 points, it assessed that the performance gap is not significant.