What do you mean? As in, they are reallocating compute away from previous models?

If so, I would think that would result in worsened performance, not quality (unless you are also suggesting they may be quantizing).

Spot on. (I had not considered quantization - that's a thought ...)