The increasing volatility and geopolitical uncertainties make accurate gold price prediction very crucial for institutional and individual investors. Gold, as a traditional safe value, plays a central role in portfolio diversification and risk management. This study presents an innovative hybrid model for the prediction of gold prices in the very short term (minutes), addressing the specific challenges of the gold market, characterized by its sensitivity to global events and high liquidity. The proposed model combines the advantages of convolutional neural networks (CNN), long-term memory networks (LSTM), and attention mechanisms (AM), then integrating their outputs into a Bidirectional LSTM (Bidirectional LSTM) layer. This architecture aims to effectively capture the complex characteristics of the gold market, including short-term trends, rapid fluctuations, and macroeconomic influences. It responds to the limitations of existing prediction algorithms, which are often insufficient in the face of the complexity and speed of market movements. The model performance was evaluated using standard metrics (MSE, MAE, R2, RMSE, MAPE, Logcosh), demonstrating very encouraging results. This significant improvement over existing models provides investors with a more accurate tool for their trading and portfolio management strategies. This research aims to contribute to the field of quantitative finance by offering a new solution for short-term gold price forecasting, meeting the specific needs of market participants in a complex and ever-changing economic environment.