The growth of the e-commerce industry in Indonesia has generated a huge volume of user comments, which contain important opinions for business people and NLP researchers. Automatic processing of comments through deep learning models poses a challenge, especially in the context of sentiment classification. This study aims to compare the performance of two transformer-based models, namely IndoBERT and IndoGPT, in sentiment analysis tasks on Indonesian-language e-commerce beauty product comments. The method used is quantitative comparative with testing two hyperparameter scenarios (variations in batch size, learning rate, and epoch), and using evaluation metrics in the form of precision, recall, f1-score, and accuracy. The main contribution of this research is to present a head-to-head evaluation of IndoBERT and IndoGPT on an informal Indonesian-language e-commerce dataset, a context that has not been directly tested in previous literature. The results of the experiment show that IndoBERT consistently provides superior results compared to IndoGPT on all sentiment labels. The highest accuracy was achieved by IndoBERT at 83 percent, surpassing IndoGPT, which only reached 80 percent. These findings indicate that IndoBERT is more effective in handling class imbalance and language complexity in product reviews, making it more suitable for application in automated opinion analysis systems on e-commerce platforms.