The paper introduces Active Generalized Category Discovery (AGCD), a novel setting that addresses the challenges of Generalized Category Discovery (GCD) by leveraging active learning. GCD aims to cluster unlabeled samples from both novel and old classes, but faces issues such as imbalanced classification performance and inconsistent confidence between old and new classes. AGCD proposes an adaptive sampling strategy, Adaptive-Novel, which considers novelty, informativeness, and diversity to select valuable samples for labeling. Additionally, a stable label mapping algorithm is introduced to transform ground truth labels into the classifier's label space, ensuring consistent training across different active selection stages. The method achieves state-of-the-art performance on various datasets, demonstrating its effectiveness in improving category discovery with limited annotations.The paper introduces Active Generalized Category Discovery (AGCD), a novel setting that addresses the challenges of Generalized Category Discovery (GCD) by leveraging active learning. GCD aims to cluster unlabeled samples from both novel and old classes, but faces issues such as imbalanced classification performance and inconsistent confidence between old and new classes. AGCD proposes an adaptive sampling strategy, Adaptive-Novel, which considers novelty, informativeness, and diversity to select valuable samples for labeling. Additionally, a stable label mapping algorithm is introduced to transform ground truth labels into the classifier's label space, ensuring consistent training across different active selection stages. The method achieves state-of-the-art performance on various datasets, demonstrating its effectiveness in improving category discovery with limited annotations.