Since there are lots of models in PROCESS, there is no absolute minimum, except 2 samples per group to be able to calculate a variance. But this not what you mean I think. The effect size of practical relevance in your research field or smallest effect size of interest (SESOI) and power and alpha determine the sample size. There is no absolute value, especially if you may have different models (mediation, moderation....), there is no easy way to determine it.
I still dont understand your question. It doesnt make sense to me. You choose your model, which fits to your hypotheses and your data generating process. If you have only 1 moderator and all assumptions are met, model 1 is fine, but if you have a mediation hypothesis this wont make any sense for example. Thefore, model 1 is not better than others.