Flowchart to derive ML-accelerated three-dimensional atmosphere models. The ML models (DNN, XGBoost) were trained to predict the gas temperature and wind structure of exoplanetary atmospheres. Step 1: a grid of pre-calculated three-dimensional planetary atmospheres (ExoRad) is used to train ML models. Step 2: the performance of the ML models is evaluated on unseen planets. Their predicted one-dimensional (T_gas, p_gas)-profiles are compared to reference ground truth (T_gas, p_gas)-profiles (ExoRad). Step 3: to assess how prediction errors propagate, equilibrium gas phase chemistry calculations (GGchem), performed on selected one-dimensional (T_gas, p_gas)-profiles, and simulated transmission spectra (petitRADTRANS) are compared.