Contaminant transport modelling has a structure that catches people out. You are running two models, not one. A flow model establishes where the water goes, and a transport model rides on top of it and moves a solute through that flow field. Choose the transport code first and you will have chosen your flow model by accident.
In practice the flow model is MODFLOW almost everywhere, and the real choice is which transport code sits on it and which environment you drive them from.
MT3D-USGS, and why it is the default
MT3D and its current USGS form handle advection, dispersion, sorption and simple decay for a single species or several that do not interact. That covers a large share of real contaminant work: a plume from a known source, a question about where it goes and how fast.
It is the right starting point when your species behaves conservatively or decays simply, and when the question is extent and travel time rather than chemistry.
RT3D, when the chemistry matters
RT3D exists for reactive multi species transport. If your contaminants degrade into one another, if you are modelling a chlorinated solvent daughter product chain, or if you are representing biodegradation with electron acceptors, the reaction module is the reason to be here.
The cost is that you now have to defend a reaction network as well as a flow field, and reaction parameters are usually the least well constrained numbers in the whole exercise.
SEAWAT, when density drives the flow
SEAWAT couples flow and transport so that the solute changes the density of the water and therefore the flow itself. That coupling is the whole point and it matters in exactly one common situation: saline intrusion in a coastal aquifer.
If you are modelling a coastal wellfield and the question is how far the interface moves under a given abstraction, this is the code. Using an uncoupled model there will give you a confident and wrong answer.
PHT3D, when you need real geochemistry
PHT3D couples transport to PHREEQC, which means the full apparatus of equilibrium and kinetic geochemistry: speciation, mineral dissolution and precipitation, ion exchange, surface complexation.
This is the right tool for arsenic mobilisation, for managed aquifer recharge where injected water reacts with the host rock, and for acid mine drainage. It is also considerably more work, and it needs water chemistry data most sites do not have.
The environments, and what they actually buy
GMS, Visual MODFLOW and FEFLOW are not alternatives to the codes above. GMS and Visual MODFLOW are pre and post processors that drive MODFLOW and its transport family, and what you are buying is grid construction, data import, calibration workflow and figures your client can read. On a large or long running project that tooling is worth real money. On a small one it may not be.
FEFLOW is the genuine alternative, because it is a finite element code with transport and density coupling built in rather than bolted on. Where geometry is complex or layers pinch out, its meshing is a real advantage, which is the same trade we set out in FEFLOW vs MODFLOW.
How to choose
Work down the list. Does density matter? Then SEAWAT or FEFLOW. Does the chemistry react? Then PHT3D if it is geochemistry, RT3D if it is degradation. Neither? Then MT3D-USGS, and spend what you saved on the flow model instead.
What actually determines the answer
Transport results are dominated by the flow field and by the source term. A beautiful reaction network on top of a flow model that has not been calibrated against heads is decoration. So is a precise plume boundary derived from a source history that was assumed rather than established.
Dispersivity deserves particular suspicion. It is scale dependent, it is routinely fitted rather than measured, and it will move your plume extent more than most of the choices above.
We work in these codes on groundwater modelling and water quality projects. If you are scoping contaminant work and are not yet sure whether your problem is really a density or a chemistry problem, that is worth settling before anything is licensed, and it is part of what our consulting and modelling services do.