Cyberbullying significantly harms the internal health of communities by negatively affecting victims psychological well-begin. It’s a current problem on SM platforms, challenging effective, real-time discovery and monitoring system to find dangerous dispatches. Despite progress, existing cyberbullying detection system still struggle with performance, dataset quality, speed and computational expense. This exploration used to conduct a relative study by conforming and assessing being ML models within the bullying discovery place.This paper evaluates the effectiveness and performance of detecting cyberbullying text on social media platforms. Machine learning techniques like GPT-2.0, BERT, XLNet, RoBERTa and DistilBERT are compared and BERT score 95% in precision, Recall and F1 score and 5% Error rate.