Has Artificial General Intelligence Actually Arrived?

The claim resurfaces every time a major model ships, and this week it came from OpenAI, whose president floated the idea that its new GPT-6 Astra could mark the arrival of artificial general intelligence. It is a phrase that carries enormous weight and very little agreed meaning, so it is worth slowing down and asking what would actually make it true.

Start with the definition, because there isn't a settled one. OpenAI tends to describe AGI as a system that can do economically valuable work as well as or better than humans. Others set the bar at flexible, human-level competence across almost any task, including ones the system was never trained for. Those are different tests, and a model can pass the first while falling well short of the second.

Capable is not the same as general

Today's best models are genuinely impressive. They write software, pass professional exams, and hold long, complex conversations. What they do not yet show is the easy transfer that defines general intelligence in people: the ability to walk into an unfamiliar problem, with no relevant training data, and work it out from first principles. Models still stumble on tasks that fall outside their training distribution, and they can be confidently wrong in ways a competent human expert would not be.

There is also a commercial tilt to watch for. A lab that calls its model AGI is making a marketing claim as much as a scientific one, and "AGI" has no benchmark you can point to the way you can with a coding score. When the word comes from the company selling the product, treat it as a hypothesis rather than a result.

Why the label matters

This is not just semantics. The AGI label shapes regulation, investment and public fear, and declaring victory too early muddies all three. It also moves the goalposts in an unhelpful direction. Every time a model clears a bar we thought meant "intelligence", the bar quietly shifts, which is itself a sign we never had a crisp definition to begin with.

The honest read on Astra is that it looks like a strong step forward, and a strong step forward is not the same thing as a general mind. We have looked before at the assumption that a bigger model is automatically a smarter one. The AGI question is the same instinct wearing a grander coat. Be impressed by what these systems can do, and keep a clear eye on what the word is being asked to carry.