By integrating a machine learning model for eddy reconstruction with an optimized objective analysis method, this study developed a daily dataset (Eddy Model Fusion Gradient Optimal Interpolation Data, EFGO) target for mesoscale eddy dynamics with a spatial resolution of 1/16° in the Pacific western boundary current region (114 °E–135 °E, 15 °N–35 °N). The iterative correction of background, quality control of multi-source data, gradient constraints on background error in the fusion method, and repeated validation of adjustable parameters collectively ensure the reliability of the dataset. More than 70 % of the theoretical analysis errors are less than 0.1, with a maximum not exceeding 0.5. Through cross-validation with other datasets, EFGO demonstrates finer feature extraction for mesoscale processes. Compared with in-situ Argo profiles, the maximum temperature and salinity biases of EFGO do not exceed 0.5 °C and 0.05, respectively, which are smaller than those of other eddy-resolving datasets such as GLORYS12 and HYCOM. EFGO aligns more closely with the observations than GLORYS12 and HYCOM along both Argo and WOD observation sections. Characterized primarily by enhanced eddy representation, this dataset includes multiple variables such as temperature, salinity, thermocline, sound channel depth, and eddy parameters. It provides crucial data support for the comprehensive and detailed description of refined marine environmental features. Furthermore, it is applicable in various fields including ecological dynamics and military oceanography.