Digital news platforms are emerging as pivotal channels for capturing public opinion, yet their analysis remains challenging, especially in linguistically diverse regions such as Morocco. In this study, we present a comprehensive investigation into sentiment analysis (SA) on Moroccan news headlines by leveraging Arabic-specific pretrained language models (PLMs) designed for both Modern Standard Arabic (MSA) and Moroccan Dialectical Arabic (MDA). In this study, we have enhance our approach by evaluating multiple models compared to our previous work, Societal Wellbeing Scoring and Monitoring System (SW-SMS), using a BiGRU model with ASAFAYA embeddings for sentiment analysis, achieving 90.23% accuracy on the Sentiment-Annotated Hibapress Moroccan News Arabic Dataset (SAHMNAD) we previously collected using web scraping of data in the year 2022. Notably, our new benchmarks show that CamelBERT-MSA (BERT-base-Arabic-CamelBERT-MSA) and QARiB achieve accuracies of 91.96% and 91.49%, respectively, significantly outperforming general-purpose multilingual alternatives. These advancements underscore the importance of domain-specific benchmarks in advancing Arabic natural language processing (NLP). By enhancing our understanding of public sentiment, our improved SA framework supports more informed policy-making, stocks forecasting, and market research studies.