The sensible flavors of "it's just a model" are ones that insist that any new model will still have to explain the data explained by the old model.

This may come as a rather different conceptualization which has, as a limiting case, something very alike the existing model (e.g. relativity, which reduces to Cartesian space and Newtonian mechanics). Or it may come as an important refinement (e.g. the existence of microRNA as an exception to the DNA -> RNA -> protein model of how the cell accomplishes things including, say, regulation of expression).

This leaves the truth status of something like kinetic energy being (1/2)*m*v^2 in a rather awkward position. On the one hand, it's exactly true to within any practical measurement error for certain masses and velocities. It really is a rule of the universe. Sort of.

CPT conservation is incredibly well-verified. It's how the universe works. Except...the universe itself seems like it has probably violated CPT given the dominance of matter over antimatter. Maybe CPT isn't quite how the universe works. Maybe it is, and we're missing something else fundamental. But regardless, the models overall can't be exactly right.

Of course, any "it's just a model" perspective that doesn't distinguish how good various models are falls afoul of Asimov's famous quote. And unfortunately, it doesn't seem that philosophers invariably avoid making that blunder.

But to some extent, any argument about realism vs. alternatives are expressions of what the most productive method is to emotionally grapple with observations. "Shut up and calculate" is, as much as anything else, a statement about pragmatic epistemology given human psychology.