At this scale, it's purely quality. The better the model, the faster the advancements. If using a model half as smart as the best made us half as productive, people would pretty much all be using the current quantized models that can run on a decent laptop. The difference between Opus xHigh and Gemma4 is very different (at least in my job).

YMMV and you underestimate their ability to run scale. They are not running on laptops.