Anyone have any luck getting local LLMs to perform well enough to actually do work with?

Would love to replace claude code, etc. and be without ai subscriptions, but so far the experience is v bad.