Thanks to the increase in resolution, numerical weather prediction models resolve a growing fraction of the gravity wave (GW) spectrum. Here, we assess the representation of Lagrangian GW fluctuations by comparing trajectories in the models to long-duration balloon observations. Most characteristics of the observed GW spectrum, such as near-inertial oscillations, are qualitatively present. However, the variability remains underestimated, emphasizing the continuous need for GW parameterizations.