New preprint: GRAM, spatial general-purpose audio representation models, with the NatHEAR and RealSELD benchmarks.