Abstract This study evaluates the performance of 75 clear-sky global irradiance models against 75 ground stations worldwide covering five major Koppen-Geiger climate classifications and overall global performance. After quality control, clear-sky detection and data-availability criteria, there are 4.36 million 1-min valid global horizontal irradiance (GHI) data points for evaluation. This study represents the most encompassing evaluation of its kind in terms of number of models assessed, number of ground stations used, most consistent selection of input variables in terms of temporal and spatial resolution, also using the most rigorous and fair performance assessment criteria and suitable ranking system. A statistically rigorous Principal Component Analysis (PCA) ranking procedure is proposed to replace the conventional ordering method based on single and simple statistics. In particular, it is demonstrated that all 13 error metrics contribute to the variance of GHI estimation and thus must all be accounted for in the ranking procedure. The best performing models in each climate zone of equatorial, arid, temperate, cold and polar are REST2v9.1, REST2v9.1, MAC2, REST2v5 and CLS, respectively. Globally, the top three performing models are MAC2, REST2v5, and REST2v9.1. Many models appear to suffer from over-fitting empirical relationships to training data, resulting in inconsistent worldwide performance. Furthermore, six different formulations of the Linke Turbidity factor (TL) are evaluated through its application to five clear-sky models that require it as an input. The best globally performing TL formulations are either by Ineichen or Gueymard when combined with four of the five Linke-dependent clear-sky models, whilst the Grenier formulation performs best with the Ineichen & Perez model. Significant performance variation is observed depending on the TL formulation selected, hence, careful selection is required for an optimal application. The model codes are available in R [1].