Google used to do that with a version number and descriptor. 2.5 Pro, 3.1 Flash-Lite, etc. Gave you both model generation + idea of relative size/quantization within generation. I guess they decided that scheme wasn't flashy enough?