GAMS flags rows and columns in the solution by NOPT or INFES if they are non-optimal or infeasible. The NOPT basically signal the marginal of the row or column has the wrong sign. A marginal on a row is simply the dual and a marginal on a column is the reduced cost (this can be interpreted as the dual if the bound was written as a constraint instead).
If the model is infeasible it does not make sense to report NOPTs. In this case we don’t even know about which objective we are talking about (phase I or phase II objective). Hence the following output of GAMS/CPlex is really a bug in the Cplex link: