SoilGrids250m is a global gridded soil information system based on machine learning, providing predictions for standard soil properties at 250m resolution. The system uses 150,000 soil profiles and 158 remote sensing-based covariates, including climate, topography, and land cover data. It employs machine learning methods like random forest and gradient boosting to improve prediction accuracy. The system includes 280 raster layers for soil properties and classes, with predictions for seven depths (0, 5, 15, 30, 60, 100, and 200 cm). The results of 10-fold cross-validation show that the ensemble models explain between 56% (coarse fragments) and 83% (pH) of variation, with an overall average of 61%. Improvements in accuracy compared to the previous 1km version range from 60 to 230%. The system includes expert-based pseudo-observations to address undersampled areas and uses a stack of covariates for higher resolution. SoilGrids250m data is available under the Open Database License and can be accessed via a web portal, WCS, and the SoilInfo App. The system is designed to provide more accurate, complete, and consistent global soil information, with potential for future improvements in incorporating input uncertainties and deriving posterior probability distributions. The system is available for download from www.SoilGrids.org and includes detailed technical documentation and code.SoilGrids250m is a global gridded soil information system based on machine learning, providing predictions for standard soil properties at 250m resolution. The system uses 150,000 soil profiles and 158 remote sensing-based covariates, including climate, topography, and land cover data. It employs machine learning methods like random forest and gradient boosting to improve prediction accuracy. The system includes 280 raster layers for soil properties and classes, with predictions for seven depths (0, 5, 15, 30, 60, 100, and 200 cm). The results of 10-fold cross-validation show that the ensemble models explain between 56% (coarse fragments) and 83% (pH) of variation, with an overall average of 61%. Improvements in accuracy compared to the previous 1km version range from 60 to 230%. The system includes expert-based pseudo-observations to address undersampled areas and uses a stack of covariates for higher resolution. SoilGrids250m data is available under the Open Database License and can be accessed via a web portal, WCS, and the SoilInfo App. The system is designed to provide more accurate, complete, and consistent global soil information, with potential for future improvements in incorporating input uncertainties and deriving posterior probability distributions. The system is available for download from www.SoilGrids.org and includes detailed technical documentation and code.