AIFS Oper is ECMWF’s AI-based global forecasting system run alongside the physics-based IFS. It replaces numerical solvers with a learned model trained on ERA5 and fine-tuned on operational IFS analyses. The network uses a graph-based encoder/decoder
with a transformer processor to predict atmospheric evolution from an analysed initial state. Inputs span pressure-level and surface fields plus simple forcings (for example time of year and insolation). Closely related suites are the physics counterpart IFS Oper and the ensemble variant AIFS Ensemble.