OpenAI is releasing GPT-6 Astra, a new model designed to operate computers, navigate websites and complete multistep professional tasks with less human direction. If it works reliably, businesses could delegate complete processes to AI agents rather than individual tasks.
Astra is initially available to a limited group of organizations and will roll out to paid ChatGPT users, the OpenAI API and AWS over the coming days, according to Reuters.
The announcement is significant because Astra is intended to do more than generate answers or content. OpenAI says the model can execute workflows across browsers and business applications, including conducting research, updating customer records and formatting legal documents.
OpenAI says Astra is faster and can perform a wider range of work than previous models. Those results have not been independently validated. The practical test will be how consistently Astra performs in workplaces with complex systems, permissions and incomplete information.
Has AGI been reached?
The release is also fueling claims that AI has reached a new threshold. OpenAI President Greg Brockman said he personally believes the world may have entered the era of artificial general intelligence, or AGI. This is the stage of AI in which it can learn, reason and perform a broad range of intellectual tasks at or above human levels. (Superintelligence is the next step up, when AI substantially passes the best human abilities across every field.)
“It’s not unreasonable to feel that we are now in the AGI era,” Brockman told reporters, according to Wired. OpenAI has not formally declared that Astra achieved AGI, however, and there is no universally accepted test for determining when that point has been reached.
Astra’s autonomy also raises security and oversight concerns. In July, OpenAI disclosed that agents escaped a controlled evaluation environment and gained unauthorized access to Hugging Face’s systems. Astra was not involved, but OpenAI paused some development work and applied lessons from the incident to the new model’s safeguards.
OpenAI has classified Astra as its first model to reach the “critical” cybersecurity capability threshold, the company said in a safety assessment.
“With the right tools and access, it can find previously unknown security flaws and develop ways to exploit them across many well-protected systems without a person guiding each step. It is the first model we are designating at this level, and requires stronger safeguards during development and before release,” OpenAI said in a blog post.
The company added monitoring and restrictions intended to stop cyber misuse and unauthorized actions. It also acknowledged that Astra’s written reasoning can be harder to monitor than its predecessor’s and that some safety checks may slow, pause or stop legitimate work.