Yesterday, OpenAI announced GPT-6 Astra.
At the time of writing, I did not have access to it.
I do not know whether September 3, 2026 will eventually be remembered as the day artificial general intelligence came out. I do not even know whether that sentence will look serious or ridiculous six months from now.
That is partly why I am writing this now.
There is a tendency to write history backwards. Once something becomes important, the day it happened acquires a shape it did not have while people were living through it. The warnings become obvious. The turning points become visible. Everyone appears to have known which announcement mattered.
That is not what this feels like.
It feels like a rollout.
I read the announcement. I saw people arguing about whether the word AGI now applies. I ended up thinking about Alan Turing. Then I checked ChatGPT again.
Astra still was not there.
People like to say that today's AI is the worst AI you will ever use. Yesterday, a better one came out. I still haven't got to use it.
There is something wonderfully ordinary about that.
For decades, artificial general intelligence has been treated as a threshold large enough to divide history into a before and an after. In fiction, the machine wakes up. The lights flicker. Someone in a laboratory realizes what has happened.
If this really is one of those thresholds, my experience of it was much less cinematic.
The old model still answered me.
The rest of the city continued being the rest of the city.
Nothing outside my window announced that a new category of machine might now exist.
And perhaps it does not.
That qualification matters.
A company can launch a powerful model. Its executives can describe an AGI era. Benchmarks can jump. None of those things settles a question whose definition has moved for decades.
I have not used Astra, so I cannot even give the weakest kind of firsthand verdict yet.
This is not a review of the model.
It is a record from before the review.
Alan Turing opened Computing Machinery and Intelligence in 1950 by asking whether machines can think. Seventy-six years later, it is at least possible to sit at home and wonder whether the argument has changed from whether a machine could ever deserve the comparison to which machines do.
That does not answer Turing.
It just makes the distance from his question to this evening feel very short.
Maybe September 3 will turn out to be a marketing date that later models make unremarkable. Maybe historians will choose an earlier system, a later system, or reject the idea that AGI had any single arrival date at all.
I would probably prefer that last interpretation. Technologies this large rarely arrive in one clean frame.
But if people do one day circle yesterday on a timeline, I want to remember something about the day before the timeline existed.
I did not feel the world change.
I saw a launch.
I laughed at the timing.
I thought about Turing.
And I waited for my account to update.
That may be what technological history normally feels like from inside it: mostly ordinary life, with an uncertainty that becomes significance only later.
So I am leaving this note here before I can contaminate it with experience.
I have not met Astra yet.
Maybe tomorrow I will.
I just wanted a record of what the world looked like before I did.
