We compare the hybrid, stochastic, and differential semantics for stochastic Concurrent Constraint Programming, focussing on the exhibited behavior of models and their robustness. By investigating in detail two case studies, a circadian clock model and the Repressilator, we comment on the effect of the introduction of a limited amount of discreteness in the description of biological systems with hybrid automata. Experimental evidence suggests that discreteness increases robustness of the models.