The formation and composition structure of soil mass, a type of geological entity, are extremely complex, and its properties exhibit typical spatial variability. Traditional Monte Carlo simulation usually employs a single random variable to simulate the variability of soil mass. A single random variable model cannot, however, accurately capture the spatial variability of soil mass since it is more tied to the soil's spatial location than the simply statistically random. In this study, the stochastic material point method is developed by combining the random field theory into the material point method. The risk of the whole landslide process has been quantified based on two quantitative evaluation indices: sliding mass and sliding distance. Moreover, the impact of various random field correlation functions (such as the triangular, Gaussian, exponential, and exponential cosine functions), different types of probability distributions (such as the Gaussian normal distribution and lognormal distribution), and various soil mass shear strength parameters are compared and examined. Results demonstrate that by taking into consideration of the spatial variability of soil mass, the stochastic material point method can more accurately assess the overall risk of a soil landslide. Compared with the traditional Monte Carlo material point method, the stochastic material point method can significantly improve the computing efficiency, which is more suitable for large-scale engineering applications. The correlation function and type of probability distribution have an impact on the simulation results, and when the trigonometric function, the lognormal distribution, and the Gaussian normal distribution are used as the correlation function, cohesion, and internal friction angle, respectively, the results are in good agreement with the Monte Carlo simulation. Furthermore, compared to cohesions, the spatial variability of internal friction angle is more sensitive to the risk of landslides.