Deeper keeps winning: contrastive RL agents gain new capabilities all the way to 1,024 layers (Wang et al., 2025).