Clouds play a crucial role in Earth’s climate system, with clear-sky albedo being fundamental for estimating cloud albedo and the shortwave cloud radiative effect (CRE), which are key to understanding Earth’s radiative balance. However, direct satellite measurements of theoretical clear-sky albedo for cloudy pixels are impossible. To address this limitation, we developed a Multi-Layer Perceptron (MLP) model trained on over 20 million samples from the Clouds and the Earth’s Radiant Energy System (CERES) dataset, enabling the estimation of instantaneous clear-sky albedo at the top of the atmosphere. The MLP model exhibits robust predictive performance, achieving an RMSE of 0.004 and R² of 0.96, demonstrating a closer agreement with direct observational products compared to other radiation products, with reliable estimations across various years and cloudy regions. Furthermore, we identify and correct undetected sub-resolution cloud contamination within clear-sky pixels present in CERES observations and the errors introduced during temporal interpolation between clear-sky pixels in the CERES Synoptic TOA and surface fluxes and clouds (SYN) product. Based on clear-sky albedo, the estimated instantaneous noon shortwave CRE is -113.44 W·m⁻². By employing another MLP model to scale the instantaneous clear-sky albedo to daily values, the estimated daily CRE is -44.51 W·m⁻², which is 1.02 W·m⁻² weaker than the estimated CRE from the CERES SYN product. Most of the correction accounts for temporal interpolation errors. The deep-learning-derived clear-sky albedo and the estimated CRE provide a new approach for research on aerosol-cloud interactions, cloud feedback mechanisms, and model improvements, offering valuable insights into the field.