I feel most tasks don't need a frontier model.

The other day I was bouncing ideas back and forth with an LLM, when I realized the effort was at high, and the model was honestly quite powerful.

That whole brainstorming session didn't need to tap into frontier intelligence to crunch some numbers like a regular calculator could, and help me navigate new concepts, in fact a decent 8B would have handled the whole thing just as I had needed anyways. YET I was using the literal frontier of human technology to do everyday stuff.

Running LLMs locally has always had some friction to it, GPT sits there in the browser, Claude, Deepseek, all of em are super accessible. That's definitely easier than installing a runtime and pick between things that require thoughts and process, just to get a model that is noticeably worse at tasks.

In my X timeline I'm really starting to see those quants becoming something people, devs and labs are running everyday locally, on their own hardware in alternative to them anthropics and GPTs.

Small models got a lot better over the past year, and they are starting to run on laptops and smaller devices. The setup is still A LITTLE annoying, which I think is the real reason most people never tried again after the first bad experience.

So I’m curious where everyone actually lands on this (what's the general consensus looking like);

If running a capable model on your own machine took one click and cost you nothing, would you move your everyday tasks over? Or is there something about them GPTs and Anthropics you’d keep going back for?