Large databases of climate model simulations are essential to sample climate variability and estimate how it can evolve in any future. The chaotic nature of climate has motivated the simulation of large ensembles of simulations, which sample the uncertainty due to internal climate variability in single models. Exploiting large ensembles (for impact or attribution studies) implicitly relies on the hypothesis that simulations are interchangeable. This is not the case for variables like temperature, due to biases (which can be corrected). Some synoptic fields, like SLP, do not yield obvious biases, which might justify their use to enrich reanalysis data. In this paper, we examine this hypothesis through a neural network classification approach. The goal is to determine whether it is possible to recognize a climate model (among 16 models and a reanalysis) from one single sea-level pressure (SLP) map over the North Atlantic. We find that models are highly identifiable in the summer (and less in other seasons), while SLP average structures are very similar. From this classification, we identify sororities of climate models, and investigate how climate change can affect SLP daily patterns toward the end of the 21st century. This study allows identifying which climate models could be used as input for artificial intelligence model forecasts.