Artificial general intelligence has no formal definition in machine learning and no agreed benchmark to certify it. OpenAI says its new model, called Astra, could be considered artificial general intelligence, and separately claims the model has overtaken Anthropic's AI systems. The ChatGPT maker has attached both assertions to the same product.
The competitive claim against Anthropic is the more tractable of the two. That Astra surpasses Anthropic's current systems is at least testable in principle, if OpenAI discloses the underlying evaluations. The AGI framing operates under a different constraint. Without a pre-specified success boundary, the designation is self-graded by whoever issues it. OpenAI's conditional phrasing, "could be considered," advances a threshold the industry has not formalized.
The weight of both claims depends on what evaluations OpenAI used to reach them.