Anatomical human models are extensively utilized for assessing induced electric fields due to low-frequency (LF) electromagnetic exposure. One difficulty in the LF dosimetry is that the results are often affected by numerical artifacts, which are attributable to the abrupt change at tissue interfaces for the segmented human models with discrete tissue conductivities. To overcome this difficulty, head models with continuous conductivities have been recently developed using deep learning networks, which directly map magnetic resonance images to volume conductivity without segmentation. To validate the effectiveness of this novel modeling method for electromagnetic dosimetry, a working group was established by the IEEE International Committee on Electromagnetic Safety Technical Committee 95 Subcommittee 6. The group’s initial study focused on intercomparison of computed fields using learning-based models across several laboratories. This paper extends the analysis considering the effect of conductivity variations on the computed electric field induced in learning-based continuous models and segmented discrete models. Six international research groups participated in this joint study. It is found that the electric field strengths decrease in grey matter (GM) and increase in white matter (WM) as GM conductivity increases. Electric field strengths in both GM and WM decrease as WM conductivity increases. The variation ranges of electric field strength, due to varying conductivity values, show comparability between discrete and continuous models. For the intercomparison, the highest relative differences (RDs) are 15.9% and 6.7% for the 100th and 99th percentile values of the induced electric fields for the discrete models, respectively, and 10.1% and 3.8% for the continuous models. The RDs for computations using the scalar-potential finite-difference method with different solvers are below 1.2%.