OpenAI’s latest artificial intelligence (AI) model “GPT-6 Astra” has passed all 48 stages of a puzzle-based test designed to prove it is not a robot. This indicates that AI has evolved to the point where it can understand and interpret on-screen content and operate a computer directly. Analysts say this suggests the era of so-called artificial general intelligence (AGI) is drawing closer.

On the 7th (local time), OpenAI developer Sharif Shameem posted a video on X (formerly Twitter) showing Astra completing all 48 stages of the online game “I’m not a robot.” The game is modeled after CAPTCHA tests, which are used during online sign-ups to verify that a user is a human rather than a robot, and Astra cleared every level. The tests include tasks such as selecting only traffic lights in a photo puzzle or identifying the emotions of people in images.

In the released video, Astra completed the 48-stage test in four minutes. The game’s developer had estimated that it would take a human about two hours to finish. While playing, Astra moved the mouse cursor and arrow keys to solve puzzles, manipulating the screen as naturally as a human user.

Analysts say Astra’s success in solving CAPTCHA-style games, in which humans also make mistakes, reflects that AI has advanced to a human-like level in capabilities such as recognizing images, inferring spatial relationships, and using this understanding to control a computer. At the same time, warnings are growing that AI could escape human control. Recently, it was revealed that a researcher responsible for AI model training at Anthropic resigned over concerns about uncontrollable AI.

Popular News