Borehole-measured soil temperatures have been routinely used to calibrate soil parameters in permafrost modeling, although they are sparse in the Qinghai-Tibet Plateau (QTP). A feasible alternative is to calibrate models using land surface temperatures. However, the quantitative impacts of various soil parameterizations on permafrost modeling remain unexplored. To quantify these impacts, two sets of soil parameters (denoted as P and P) were obtained via calibration using borehole temperature measurement and ERA5-Land (the land component of the fifth generation of European Re-Analysis) skin temperature, respectively, and applied to the Geophysical Institute Permafrost Laboratory Version 2 (GIPL 2.0) model. Comparing against the borehole-measured soil temperatures of 4 soil layers, the ERA5-P simulation (with Root Mean Squared Error, i.e., RMSE from 1.4 °C to 3.9 °C) outperform ERA5-P simulation (RMSE from 1.4 °C to 3.9 °C) during 2006–2014. The obtained P and P were then utilized as soil parameters in GIPL 2.0 to model permafrost dynamics for a long period from 1983 to 2019, respectively, using ERA5-Land as forcing data. Simulations revealed significant disparities. In comparison to the simulation using P results using P show that the mean annual soil temperature at 1 m depth was 2.72 °C lower with a 0.01 °C/a (50.0%) lower trend; the active layer thickness was 0.81 m (35.7%) less with a 2.16 cm/a (82.1%) lower trend; the duration of the thawing season at 1 m depth was underestimated by about one month, and the zero-curtain period was about 23 days (37.7%) shorter. The change rates of the zero-curtain period, however, were comparable. This study implies that choosing soil parameterizations is critical for model evaluation against observations and long-term model prediction.