In the extreme conditions of inertial confinement fusion experiments, heat flow plays a vital role, but local diffusive models frequently break down and overestimate the heat flow. The situation becomes more complicated again in the significant magnetic fields generated during laser-plasma interactions or in magnetized fusion schemes. Accurate non-local and magnetized heat flow computations can be carried out using Vlasov-Fokker-Planck (VFP) simulations, but these are computationally expensive. There is, therefore, significant interest in using faster multi-group models to accurately calculate the non-local heat flow in magnetized plasmas. We benchmark two such multi-group models for calculating the heat flow, M1 and hybrid-AWBS-BGK, against diffusive models and full VFP simulations, before applying the models to realistic example test cases, both magnetized and unmagnetized. We find that the multi-group models generally perform very well for moderate non-localities up to k λ mfp ∼ 0.01 , but the computational cost increases dramatically. hybrid-AWBS-BGK performs more effectively than M1 at high non-localities, up to k λ mfp ∼ 1 , due to its adaptive solver and robust P1 closure, but tends to fail in very strong magnetic fields. Both codes are much faster than VFP simulations but are still slow in steep temperature gradients.