GPT-6 Astra Is Here—and OpenAI Thinks It May Kick Off the AGI Era

OpenAI has unveiled GPT-6 Astra, a powerful new AI model that the company believes marks the beginning of the artificial general intelligence (AGI) era. The advanced model is set to be released to paid customers, including enterprise clients in OpenAI's Daybreak early access program, as well as subscribers to ChatGPT Plus, Pro, and Business plans. It is unclear if GPT-6 Astra will be accessible to free users of ChatGPT.

During a press briefing, OpenAI executives expressed confidence in the AI model's capabilities. President and co-founder Greg Brockman stated that he personally believes the world is now in the AGI era, suggesting that GPT-6 Astra is a key factor in this development. He believes that in a few years, when looking back at the creation of AGI, this model and the time it was released may be considered pivotal.

OpenAI is releasing GPT-6 Astra at a crucial time for the company, as it seeks to rapidly advance its technology and boost sales ahead of a planned initial public offering (IPO) while maintaining a strong focus on safety and security. The AI maker faces intense competition from Anthropic, another leading player in the AI space, which is also preparing for an IPO.

In a blog post, OpenAI described GPT-6 Astra as "the world's best computer use model," highlighting its ability to navigate computers and web browsers on behalf of humans more effectively than previous versions.

Tests conducted by OpenAI demonstrated that GPT-6 Astra could perform tasks such as booking DMV appointments, searching for job listings, and apartment hunting with greater speed and accuracy than the average person. The company asserts that GPT-6 Astra is also the most secure and aligned AI model released by OpenAI to date. Chief scientist Jakub Pachocki emphasized that the firm is closely monitoring GPT-6 Astra's chain-of-thought, the internal scratchpad used by the model to solve problems, to ensure its safe deployment.

However, Pachocki acknowledged that monitoring advanced AI models for unintended harm is becoming increasingly challenging and may pose a bottleneck to further AI progress. He noted that as models become more capable, monitoring their alignment becomes more difficult, and OpenAI may need to scale back development if they cannot maintain sufficient visibility into model alignment.