OpenAI launches GPT-6 Astra, and one founder thinks AGI is here

OpenAI has launched the GPT-6 Astra, its latest AI model and what the company describes as the most capable model it has used to date. Astra brings significant revenue in areas such as software development, computer exploitation and cyber security, and it will be rolled out to paid ChatGPT users and APIs in the coming week. Although OpenAI is not officially called this AGI, one of its founders believes that we have now reached this point.

as reported TechCrunchOpenAI is particularly touting Astra’s ability to use computers and browsers, and president Greg Brockman has described it as the company’s most intelligent model yet. The company also considers the Astra the best model for software development, with testing showing improvements in areas such as bug detection and working with codebases.

The unnerving factor is that Astra is the first OpenAI model to reach the “Critical” level for cybersecurity capabilities under the company’s readiness framework. OpenAI says that given the right tools and access, Astra can find previously unknown security flaws and develop ways to exploit well-protected systems without requiring a human to manage every step. This is very useful for finding vulnerabilities before attackers do, but the downside is that it may not require much explanation.

According to OpenAI, Astra GPT-5.6 is more resistant to jailbreak attempts than Sol, and its more extensive security testing generally shows less signs of problematic behavior. Given Astra’s powerful capabilities, the company applies additional controls to device-based sessions as an additional layer of protection.

When it comes to keeping track of what Astra is doing, things continue to be unpleasant. TechCrunch Emphasis is placed on the model’s use of a reasoning technique called fuzzy iteration, which makes its train of thought difficult for researchers to follow. Jakub Pachocki, chief scientist at OpenAI, acknowledged that as models become more sophisticated, control becomes more difficult, in part because capable models can handle difficult tasks using fewer language tokens, or sometimes none at all. OpenAI system map This supports a broader concern by finding that Astra’s tracking ability is reduced relative to Sol.

This brings us back to AGI. When asked on a media call if Astra signals his arrival, Brockman said AGI is no longer a contractual trigger for OpenAI and has become a “mission concept or spiritual concept.” He stopped short of declaring Astra an AGI on behalf of OpenAI, but his personal view was somewhat ambivalent: “Personally, I think we’re there.”