This study proposed a 20 m soybean yield dataset in Northeast China (NortheastChinaSoybeanYield20m) from 2019 to 2023 using a hybrid framework coupling crop growth model with deep learning algorithm. Stable results were achieved through the years. The overall accuracy of the dataset was 287.44 and 272.36 kg ha^–1 in the root mean squared error for field and regional scale, respectively. The study satisfied the urgent demands for precise control of crop yield information.