> I wonder how models perform on finding analogies between analogies

Load-bearingly verbose, in my experience.