Good read. My two cents: I am using it because it supports mlx explicitly. llama.cpp is still metal, so it's a bit slower. Although I love llama.cpp and was using it natively.

LM Studio - I don't like it, many bugs, closed source.

What I want: FOSS, models are introduced and supported fast, ideally one command (ollama pull-like) model installation, embedding models support.

I am not saying ollama is perfect, but works for me better than alternatives I looked at.