Comment by cyanydeez

When was the last time you tried a local model? You might actually be addicted to a dumb model that's just very quick at everything, which is what the GP is positing.

I've got the QWEN3.5 coder running and it's dumb, but if the task is 'clone this component and rewrite it to testWhatever' it does it perfectly well.

One might thing you're falling for the delusion being posited in this thread and you don't have a proper benchmark. You're just getting that token hit.