> The problem is that the big cloud models got better too—including those open weight models that, while freely available, were far too large (100B+) to run on my laptop.

The actual, notable progress will be models that can run reasonably well on commodity, everyday hardware that the average user has. From more accessibility will come greater usefulness. Right now the way I see it, having to upgrade specs on a machine to run local models keeps it in a niche hobbyist bubble.