Accurate forex price prediction is challenging due to the market’s inherent volatility and the complex interactions among various economic factors and market sentiment. This study addresses these challenges by integrating sentiment analysis with advanced deep learning and machine learning architectures within a multi-model framework. Over two thousand financial news headlines were collected from leading economic websites using web scraping tools such as Selenium and BeautifulSoup, forming the basis for sentiment analysis. The sentiment data was processed using cutting edge natural language processing techniques, including Zero-Shot learning with GPT-4 and GEMINI Advanced models, selected for their proficiency in handling unlabeled financial sentiment data. The proposed solution combines sentiment scores with technical and fundamental market indicators, feeding these into diverse predictive models, including Long Short-Term Memory networks, eXtreme Gradient Boosting, and Transformer architectures in encoder-decoder and decoder-only configurations. This approach captures the nuanced effects of market sentiment within a four-hour trading window, aligning closely with real-time market dynamics. The key findings indicate that integrating sentiment analysis significantly improves prediction accuracy, with the XGBoost model demonstrating superior performance when combined with technical, fundamental, and sentiment data. The study reveals that while this multi-model approach offers improved predictive capabilities, it also faces limitations such as dependency on high-quality sentiment data and the computational intensity of training complex models. These results suggest that the integration of sentiment analysis provides a competitive edge in forex forecasting, though further research is needed to refine these methods and address their limitations.