OpenAI has released its next-generation model, GPT-6, calling it a critical milestone in the company's decade-long push toward artificial general intelligence (AGI). On Thursday, the company said it has begun rolling out GPT-6 Astra to enterprise users participating in the Daybreak program, which gives approved testers access to some of OpenAI's most capable systems.

The company also plans to introduce a version of GPT-6 for paying ChatGPT subscribers, with additional safeguards designed to block access to its most cutting-edge cybersecurity features. Both versions come with built-in protections aimed at preventing users from tapping into the model's most advanced cyber capabilities.

During a press briefing, OpenAI President Greg Brockman said he believes future observers may look back at this new model as an early form of AGI — software that can outperform humans across a broad range of tasks. While AGI has become a major objective for the AI sector, there is still no universally accepted definition of the term, and skepticism remains widespread.

OpenAI and rival Anthropic have been locked in a fierce race to build increasingly sophisticated AI systems, all while navigating growing safety concerns, particularly following a string of security incidents involving both companies' technologies. In response, OpenAI said it has strengthened the new model's safeguards to reduce the risk of misuse, especially in areas related to cybersecurity.

According to OpenAI, the new AI model is especially adept at taking over computer tasks on behalf of users, handling a wider range of work-related activities and sustaining longer, more complex operations than previous versions.