Introducing GPT-6 Astra: OpenAI Believes It Could Mark the Dawn of AGI.

OpenAI revealed on Thursday the debut of its advanced AI model, GPT-6 Astra, which the company asserts excels at navigating computers and web browsers, coding software, and tackling complex mathematical problems.

The plan is to offer GPT-6 Astra to paid users in the upcoming days, initially to a limited number of organizations, including enterprise clients in its Daybreak early access initiative, followed by subscribers of ChatGPT Plus, Pro, Business, and Enterprise plans. OpenAI has not clarified whether GPT-6 Astra will be accessible to free ChatGPT users.

During a press conference with journalists, OpenAI executives expressed strong confidence in the new AI model. President and cofounder Greg Brockman stated that he believes we have now entered the era of artificial general intelligence, where AI systems demonstrate intelligence surpassing that of humans.

“It’s reasonable to think we have entered the AGI era,” Brockman shared with reporters. “If we look back a few years from now and ask, ‘When did AGI truly emerge?’ I believe we will pinpoint this moment, and this model may be key to that.”

The release of GPT-6 Astra comes at a crucial time for OpenAI. The company is eager to rapidly enhance its technology and increase sales ahead of a planned initial public offering while ensuring safety and security are not compromised. It also faces intense competition from Anthropic, its primary competitor, which is preparing for its own IPO.

In a blog entry, OpenAI refers to GPT-6 Astra as the “finest model for computer use in the world,” claiming it can navigate computers and web browsers on behalf of users far more effectively than earlier versions. Historically, AI companies have found it challenging to create models capable of reliably browsing websites, completing forms, and using software tools quickly and accurately enough to replace human users. In their testing, OpenAI indicates that GPT-6 Astra was able to schedule DMV appointments, search for job listings, and find apartments more swiftly than the average individual.

The company also asserts that GPT-6 Astra is the safest and most aligned AI model they have produced to date. Jakub Pachocki, OpenAI’s chief scientist, informed reporters that they are tracking GPT-6 Astra’s chain-of-thought—its internal process for problem-solving—to guarantee safe deployment. However, Pachocki acknowledged that ensuring advanced AI models do not inadvertently cause harm is an increasingly difficult task, which could hinder further advancements in AI.

“We often take this monitoring capability for granted, but as model capabilities grow, monitoring becomes more complex,” Pachocki explained. “We believe that confidence in our monitoring could limit future development, as we cannot risk our ability to monitor model alignment declining beyond a certain threshold. We will delay scaling until we can restore sufficient confidence.”