Researchers released AMARO v2.0, a coarse‑grained neural‑network potential that keeps every heavy atom while treating solvent implicitly. The model introduces hydrogen‑aware bead types and a total‑charge conditioning scheme that lets a single representation cover proteins, drug‑like molecules, and their ternary complexes. Across 262 biomolecular complexes covering 11 targets, AMARO v2.0 preserves the folded state of proteins and recaptures most ligand poses. In a test involving the HIF2A ternary complex, the model correctly reoriented the quaternary structure. In focused studies, AMARO v2.0 distinguished two protonation states of trypsin‑benzamidine and reached the crystal binding pose in 3.3‑fold more simulation trajectories than traditional all‑atom molecular mechanics when run under the same estimated GPU‑time budget. However, the framework still struggles with binding‑pose recovery for SHP2 and site specificity for CDK8, indicating remaining limitations. Despite these caveats, the work demonstrates that a shared heavy‑atom potential can reliably simulate the dynamics of diverse biomolecular complexes.