Crop simulation models are indispensable tools that facilitate studies involving yield impacts, adaptation and management strategies, and policy analysis. However, uncertainty among models with respect to climate change factors varies with location according to genetic, site and management differences. This study aims to evaluate the accuracy of three process-based soybean models, GLYCIM, SoySim, and DSSAT CSM-CROPGRO-Soybean, which vary in representation of soil-water-plant components, with respect to soybean growth, development, and seed yield for multiple varieties and at different locations under current and projected climates. Experimental data from the United States Mississippi Delta region including 156 site-year-cultivar-irrigation combinations were used. Sensitivity analyses with respect to climate change scenarios involving increasing air temperature and CO2 on yield were also conducted using mean model ensemble values. Statistical criteria for estimating the goodness-of-fit of the models were Wilmott's index of agreement (IA) and root mean square error (RMSE). Simulated seed yield RMSE across all validated datasets was 0.92 Mg ha−1 or lower, with GLYCIM and CROPGRO exhibiting the best values. A similar pattern was observed for IA, which ranged between 0.82 and 0.65. Simulated yield in response to climate change factors increased by 8.8% per 100 ppm CO2 on average, and declined 4.8% per °C. Although both GLYCIM and CROPGRO models simulated yield reductions between 0.4%, and 7.2%, the SoySim model predicted positive impacts (+3.7% and + 12.1%) under two climate scenarios (∆T = 1.5 °C, CO2 = 423 ppm; ∆T = 2.0 °C, CO2 = 478 ppm). Given the difference in model structure and predictions, a multi-model ensemble approach is recommended to assess global crop production under future climate conditions along with continued improvement in model heat stress sensitivity and CO2 response. However, model improvements are still required to improve accuracy with regards to genetic x management x environmental interactions.