Recent advancements in Machine Reading Comprehension (MRC) models have sparked interest in the field of Named Entity Recognition (NER), where entities are extracted as answers of given queries. Yet, existing MRC-based models face several challenges, including high computational costs, limited consideration of entity content information, and the tendency to generate sharp boundaries, that hinder their generalizability. To alleviate these issues, this paper introduces CREAM, an enhanced model leveraging Concise query and REgion-Aware Minimization. First, we propose a simple yet effective strategy of generating concise queries based primarily on entity categories. Second, we propose to go beyond existing methods by identifying entire entities, instead of just their boundaries (start and end positions), with an efficient continuous cross-entropy loss. An in-depth analysis is further provided to reveal their benefit. The proposed method is evaluated on six well-known NER benchmarks. Experimental results demonstrate its remarkable effectiveness by surpassing the current state-of-the-art models, with the substantial averaged improvement of 2.74, 1.12, and 2.38 absolute percentage points in Precision, Recall, and F1 metrics, respectively.