Social networks are central to communication in the contemporary world. However, the anonymity afforded to users has emboldened the spread of online hate speech (HS) and poses a significant challenge. Online hate speech detection remains a formidable challenge in the digital age, particularly for languages with fewer computational resources like Norwegian. This paper presents a comprehensive study on HS detection utilizing transformer-based models designed for the Norwegian language, enhanced by hyperparameter tuning and regularization. Our investigation not only benchmarks the effectiveness of various models, with Nor-BERT_base emerging as notably superior for smaller datasets but also integrates explainable AI techniques to unveil the model’s decisions using LIME. This approach enhances the transparency of AI-driven content moderation and paves the way for future enhancements involving multilingual capabilities and advanced interpretative models.