Crop models have been used as an important tool to evaluate the agricultural response to climate conditions. This study aims to calibrate, and validate the CSM-CERES-Sorghum model and to investigate the vulnerability of sorghum yield for current (1982–1999) and future (2047–2064) epochs, by applying weather observations and climate outputs based on ECHAM, CCCma and GFDL models. Field experiments have been conducted in the experimental area of Janaúba and Sete Lagoas located in Minas Gerais State, Brazil. It has been found that the CSM-CERES-Sorghum model reasonably simulates crop phenology, crop biomass production, leaf area and yield components that are crucial to ensure the model reliability to reproduce in situ conditions. Comparison between the CSM-CERES-Sorghum results driven by the climate models and baseline observations shows that the ECHAM better reproduces the current observations. However, inaccurate results are found by utilizing the GFDL climate primarily due to lower precipitation values. This is found for both cities. Turning to future conditions, the simulations indicated that in Janaúba the average yields for current and future climate conditions were not statistically different, but in Sete Lagoas, there was a statistically significant increase in the sorghum productivity in the latter scenario. Moreover, it has been found that the simulations using the 52 sowing dates indicated that climate change modifies the grain yield projecting a delay in the most favorable planting date. According to the results the seeding of sorghum will very likely be held later in both cities.