Correct. People claim these models "saturate" yet what saturates faster is our ability to grasp what these models are capable of.

I, for one, cannot evaluate the strength of an IMO gold vs IMO bronze models.

Soon coding capabilities might also saturate. It might all become a matter of more compute (~ # iterations), instead of more precision (~ % getting it right the first time), as the models become lightning speed, and they gain access to a playground.