Submesoscale processes as well as internal waves (also called as the pulses of the ocean) of the ocean with horizontal scales from hundreds of meters to a few kilometers , take a key role in the oceanic dynamics and water mass/heat transport, and thus representing/predicting them correctly is essential to an accurate forecasting of the ocean state. However, they are difficult to be captured/predicted by the conventionally numerical ocean models due to their high nonlinearity in physics and the large requirement for computational resource. Here we demonstrate for the first time, to our knowledge, that a data-driven air-sea full-coupling regional forecast model built on a Swin-Transformer framework integrated with a Mixture-of-Experts (MoE) system, named “Volador 1.0”, has the capability of capturing/predicting them with only a few seconds for a 72-h prediction, and produces an energy spectrum well representing sub‑ to mesoscale energy cascade as expected by the classical turbulence theory, beyond its good performance in the forecasting for conventional marine variables. Our study highligts the feasibility of oceanic refinement forecasting down to submesoscale with a very light computational cost through using deep neural network, which could help improve the capability of disaster prevention and mitigation in coastal regions.