If you are looking at open models, check out Pi, as its very extensible and comes with a sane default. Maybe even roll your own.
Most harnesses (claude, codex, opencode etc.) assume that you use a cloud model. There’s no sense of optimization or finer control.
I've been running pi.dev + codex (GPT-5.4) for a long time for my workhorse stuff.
Actually tried caveman mode yesterday and it made everything SO MUCH BETTER. GPT-5.4 has a habit of being extremely verbose to a ridiculous degree, it's like it's writing a report for a CTO or something and padding everything as much as possible to sound smart.
With caveman it just gives me lists of stuff in a compact format. Perfect.