Using a spa.State will store the Semantic Pointer in an additional network of neural ensembles. Thus, probing the output of such a state will have additional noise. Thus, I would use a node to probe the ensemble. If you are only probing a single ensemble (and not a multiple ensembles as used by spa.State), you can even probe the ensemble directly. Probing the input of a spa.State is equivalent to probing a with node (as the input is a node), but you don’t really need the rest of the spa.State (i.e. you’re are using computational resources to simulate neurons that are irrelevant).
Aside: maybe I overlooked it, but it seems that in the example above model.inp isn’t connected to anything. Also note that nengo.spa is deprecated in favor of the nengo_spa module. You find more details on why and how switch in the documentation.