FWIW, I'm using the VertexAI API rather than running an LLM all the time. They have data privacy in the ToS, so I'm not worried about them training on my data. It's far cheaper and better than running a lower quality model myself. When I get around to some fine-tunings, they have options, but you can get pretty far with prompts, RAG, and agents
I personally get a lot of mileage out of an M3 Max with 36gb memory.
best is hard to definite so it depends on what you and your needs.