You're not wrong, but you are speaking from the perspective of a single implementation stack, i.e. modern Ethernet and the modern Internet.

The model is still solid, even if the most common implementation has melted a bit.

I disagree. The model was supposed to be universal but it fell apart as soon as people started doing anything slightly unanticipated with it. It's simultaneously both overengineered and fragile, not providing enough insight to justify its complexity - rather like OSI itself. There's a reason their protocol suite was a failure in the real world.

That's fair and I agree that the OSI model is not useful for a precise understanding of deployed networking technology.

I do think it is useful as an architectural device or a conceptual design goal -- i.e. a model to model your models on. :)

But I also concede that part of its teaching value is that it is a failure in practice.

It was a formalization of the ad hoc (successful!) design strategies of early networking. I see echoes of it everywhere, most obviously in the Linux kernel, and I think it's valuable for that.