Financial sentiment analysis plays a crucial role in understanding how investor sentiment, news coverage, and social media discussions impact financial markets. As market decisions are increasingly driven by information, accurately capturing sentiment from financial texts becomes essential for making informed predictions. However, the specialized language and subtle nuances of financial texts pose significant challenges for traditional sentiment analysis methods, often leading to misclassifications. In this work, we propose FinRoBERT-FSA, a fine-grained sentiment analysis approach that leverages the strengths of FinBERT, a domain-specific language model, and RoBERTa, known for its deep contextual understanding. By combining these two models, FinRoBERT-FSA captures subtle variations in sentiment while addressing the complexities of financial terminology. Experimental results on the Financial PhraseBank dataset demonstrate that our model achieves 97% accuracy, significantly outperforming traditional approaches like TF-IDF, Word2Vec, and lexicon-based methods. These findings underscore the effectiveness of model fusion in enhancing sentiment analysis for the financial domain.