We perform a qualitative analysis of performance of XPDNet, a state-of-the-art deep learning approach for MRI reconstruction, compared to GRAPPA, a classical approach. We do this in multiple settings, in particular testing the robustness of the XPDNet to unseen settings, and show that the XPDNet can to some degree generalize well. 2 Main findings XPDNet, a state-of-the-art deep learning approach for MRI reconstruction, can generalize well when compared to GRAPPA on unseen settings.