If you're a g*mer or do any work with AI (preferably open-source), pretty much anything relatively recent with a graphics card, used is better for driving down price while still getting a good deal.
even the huge models that you can't run locally are kinda mid and hallucinate a lot, why would you run some tiny local thing unless you're generating banned material
Links to what, Boots Riley's film or an LLM that will rewrite messages for you?
Because pretty much any of the open models will do the task now, depends on what hardware you have available as to what is best so I can only recommend searching "offline LLM".