Local models are astoundingly slow compared to online ones and you can’t really run anything else while they are doing inference.