The objective of this study was to carry out a reliability generalization meta-analysis of the Internet Gaming Disorder Scale (IGDS) to find out if it presents adequate values that justify its application in its four versions: original and abbreviated with dichotomous or ordinal response. A systematic review including 26 studies that apply this instrument from 2015 to June 2022 was conducted. For each version, a reliability generalization meta-analysis was performed with the random coefficients (RC) and varying coefficients (VC) model. Results showed reliability levels > 0.80 in the ordinal versions (IGDS9P and IGDS27P) and in the dichotomous 27-item version (IGDS27D), while the dichotomous 9-item version (IGDS9D) produced levels > 0.70. High heterogeneity was found in all versions (I2 > 95%; Rb > 90%). Cronbach's means with both models (RC-VC) did not differ significantly except for the IGDS9D version. It is concluded that, considering the dependence of the results on sample size, abbreviated versions do not guarantee that reliability remains acceptable, and dichotomous versions may provide limited but acceptable reliability. Finally, IGDS27P version is recommended in contexts where high precision is required.