ABSTRACT In the near future, ultra deep observations of galaxy clusters with Hubble Space Telescope or James Webb Space Telescope will uncover 300–1000 lensed multiple images, increasing the current count per cluster by up to an order of magnitude. This will further refine our view of clusters, leading to a more accurate and precise mapping of the total and dark matter distribution in clusters, and enabling a better understanding of background galaxy population and their luminosity functions. However, to effectively use that many images as input to lens inversion will require a re-evaluation of, and possibly upgrades to the existing methods. In this paper, we scrutinize the performance of the free-form lens inversion method grale in the regime of 150–1000 input images, using synthetic massive galaxy clusters. Our results show that with an increasing number of input images, grale produces improved reconstructed mass distributions, with the fraction of the lens plane recovered at better than $10{{\ \rm per\ cent}}$ accuracy increasing from $40\!-\!50{{\ \rm per\ cent}}$ for ∼150 images to $65{{\ \rm per\ cent}}$ for ∼1000 images. The reconstructed time delays imply a more precise measurement of H0, with $\lesssim 1{{\ \rm per\ cent}}$ bias. While the fidelity of the reconstruction improves with the increasing number of multiple images used as model constraints, ∼150 to ∼1000, the lens plane rms deteriorates from ∼0.11 to ∼0.28 arcsec. Since lens plane rms is not necessarily the best indicator of the quality of the mass reconstructions, looking for an alternative indicator is warranted.