Little inquiry has explicitly addressed the role of action spaces in\nlanguage-guided visual navigation -- either in terms of its effect on\nnavigation success or the efficiency with which a robotic agent could execute\nthe resulting trajectory. Building on the recently released VLN-CE setting for\ninstruction following in continuous environments, we develop a class of\nlanguage-conditioned waypoint prediction networks to examine this question. We\nvary the expressivity of these models to explore a spectrum between low-level\nactions and continuous waypoint prediction. We measure task performance and\nestimated execution time on a profiled LoCoBot robot. We find more expressive\nmodels result in simpler, faster to execute trajectories, but lower-level\nactions can achieve better navigation metrics by approximating shortest paths\nbetter. Further, our models outperform prior work in VLN-CE and set a new\nstate-of-the-art on the public leaderboard -- increasing success rate by 4%\nwith our best model on this challenging task.\n