A Nature paper introduces a machine-learning post-processing wrapper that corrects systematic forecast model errors to improve 2-to-6-week subseasonal prediction accuracy. In ECMWF's 2025 real-time forecasting competition, the tool achieved the most accurate global forecasts across all weather variables and forecast periods, beating numerical models from six operational centers, an international model ensemble, ECMWF's own AI model, and 34 competing teams' systems.