Why are sub-15B open-weights AI models becoming so rare?

A discussion on online platforms highlights that frontier AI labs have largely shifted their focus away from releasing state-of-the-art open-weights models under 15 billion parameters. Instead, labs appear to have settled on a minimum size range of around 24 to 27 billion parameters for their smaller offerings. This trend highlights the industry's prioritization of capabilities unlocked by larger scales over the edge-device efficiency that smaller models provide. Developers and consumers relying on local hardware with strict resource constraints may find fewer high-performance options available. Community members observed a clear gap where modern architectural advancements and training techniques are rarely being applied to sub-14B models. This leaves users wondering when the next breakthrough model in the smaller weight class will emerge from top research labs.