Well an on-device model is private and solves the data concerns, so it is much better

Hell, I’d might use it if it uses something like llama.cpp