In today’s digital culture, sexism and misogyny on online platforms have grown to be serious issues. To solve these problems, efficient automated sexist content detection and classification techniques must be created. In this study, we investigate the application of the GPT-2 model, a cutting-edge pre-trained language model, to the shared Exist 2023 job of sexism categorization. On the Exist 2023 dataset, we fine-tuned the GPT-2 model by adding adjustments like a classification head and weighted cross-entropy loss to tackle class imbalance. Our experimental findings show the GPT-2 model’s potential for precisely recognizing and classifying instances of sexism. Using the official assessment measure ICM (Information Contrast Measure), we assess our strategy while taking into account various evaluation modes, such as hard-hard, hard-soft, and soft-soft. The results show how well the GPT-2 model handles the problem of sexism categorization, assisting in the creation of automated techniques for fostering a safer and more welcoming online environment.