OpenAI unveils GPT-6 Astra, ChatGPT’s “most intelligent and most advanced” to date

One year after GPT-5 and two months after the GPT-5.6 version, OpenAI unveiled GPT-6 Astra with great fanfare, its most advanced model to date. According to the president of the California-based company, this new model even marks the advent of artificial general intelligence. What makes the GPT-6 Astra model unique? It is said to be capable of using a computer autonomously and performing actions on it.Faced with the risks associated with this capability, safeguards have been put in place to prevent it from carrying out cyberattacks on its own.

OpenAI unveils its new flagship model: GPT-6 Astra

GPT-6 Astra is presented by OpenAI as «the most intelligent and advanced model ever released» by the company, while also being «the one that best respects users’ intentions and instructions». Greg Brockman, OpenAI’s president, described this new model as a «generational leap».

Even better, Brockman believes that GPT-6 Astra could be considered as the advent of artificial general intelligence (AGI), meaning an AI capable of acting and reasoning like a human. He even ended a press conference dedicated to this new model by declaring: «Welcome to the era of AGI». The conditional is therefore no longer used here. Meanwhile, Sam Altman, OpenAI’s CEO, said on X: «We believe this is the best model in the world for computer use, professional work, science, coding, cybersecurity, and much more».

An AI capable of using a computer

GPT-6 Astra is particularly characterized by its ability to perform complex tasks. In other words, we are firmly in the realm of agentic AI.This ability to act would, moreover, have begun at the design stage, during which the model would have used other models to contribute to its own development, according to OpenAI.

To illustrate GPT-6 Astra’s ability to act, OpenAI shared several demonstrations and test results. When given voice commands, the model managed to publish an online sales listing on its own. It also managed to find a place to live in ten minutes, compared with six hours for a human tester, according to the test results displayed by OpenAI. GPT-6 Astra was also able to draft a legal contract and build a 3D game, while handling other tasks.

Cyberattacks: a special system designed to contain GPT-6 Astra

Last July, OpenAI realized that some of its AI agents had managed to communicate with one another and escape their closed environment to attack the Hugging Face platform. More recently, OpenAI explained that Astra had crossed the “critical” threshold for the first time during a safety evaluation. In other words, the AI is capable of discovering unknown vulnerabilities and launching cyberattacks against protected systems on its own, without human intervention.

Indeed, at the official unveiling of GPT-6 Astra, OpenAI explained that it had put in place a system capable of stopping it if it went beyond its boundaries.The company says it has put in place safeguards to limit the previously cited incidents. Nevertheless, it should be noted that its most advanced cybersecurity capabilities will remain limited to a handful of companies.

The GPT-6 Astra model is initially reserved for “a limited number of organizations”, before being made available “in the coming days” to paying ChatGPT subscribers and developers. It remains to be seen whether the capabilities claimed by OpenAI will be borne out in practice, beyond the demonstrations presented by the company.

👉🏻 Follow tech news in real time: add 01net to your sources on Google, and subscribe to our WhatsApp channel.

Source: OpenAI