Building integrated photovoltaic (BIPV) systems provide an opportunity for renewable energy generation in the built environment. In order to quantify the BIPV potential, numerical models of varying levels of complexity have been developed. This paper investigates how the complexity of BIPV models affects their predictions. The study starts with a detailed multi-physics BIPV model that combines a high-resolution one-diode model with physics-based thermal and airflow models. Next, simplifications are introduced into the model. The model predictions are compared to experimental data from a BIPV curtain wall installed in a test building in Leuven, Belgium. The results show that the detailed BIPV model is capable of estimating the BIPV daily energy yield with an average difference of 6.2 % (2.0 % for clear sky days) and the back-of-module temperature with an average difference of 1.74 °C. The use of a linear power model instead of a high-resolution one-diode model affects the average differences, but not significantly: 8.7 % for daily energy yield predictions (4.5 % for clear sky days) and 1.71 °C for temperature predictions. The use of two different empirical temperature correlations instead of a physics-based approach increases the average temperature difference to 3.5 and 4.4 °C. The average difference in daily energy yield increases to 10.2 and 10.4 %, respectively (5.9 and 5.5 % for clear sky days). These findings indicate that the detailed version of multi-physics BIPV model provides the best agreement with experimental data, but it is still possible to reduce the model complexity with acceptable accuracy.