This work presents a comprehensive evaluation of multiple word embedding models for natural language processing (NLP) applications. We begin by examining popular models and discussing desirable properties for both representations and evaluation methods. Our focus then shifts to analyzing six prominent models (BOW, TF-IDF, Word2Vec, fastText, GloVe, ELMo) through various intrinsic evaluators independent of specific downstream tasks. The results reveal that distinct evaluators capture various aspects of model quality, and some exhibit stronger correlations with real-world NLP performance. Additionally, we investigate the consistency of these intrinsic evaluators, offering insights into their complementary roles in assessing model effectiveness. This work provides valuable guidance for NLP practitioners, enabling them to select and refine word embedding models based on their specific task requirements and desired linguistic properties. This contributes to the development of more accurate and effective NLP applications across various domains.