Turing, RAM, or Von Neumann — which theory/model is the consensus?

It depends on the usage. For basic formal complexity theory, we usually work with Turing machines. However, RAMs and Von Neumann architectures are a better description for a real-world computer, so some may be more interested in them.

From my personal experience, TMs are the most popular in a CS degree, and in theory in general. I barely saw theoretical usages for RAM, and I never saw any Von Neumann architecture used in theory for computation, but I have seen models that describe a CPU in a specific course that covers that.