Gold is very popular investment instrument due to its annual prices increases. In the long term, gold prices follow a nonlinear pattern, but in the short term, there are fluctuations influenced by various factors, including global market dynamics, monetary policy, and overall economic conditions. Therefore, predicting gold prices is an important step in minimizing risk and maximizing profits for investors. In this study, we analyze the performance of two methods for forecasting global gold prices, namely long short term memory (LSTM) and autoregressive integrated moving average with exogenous variables (ARIMAX). Data used is weekly global gold price data from August 1, 2000, to June 1, 2024. The variables used are Close as the dependent variable and Open as the exogenous variable. The data used is stationary data through the differencing process and algorithmic transformation to overcome non-stationarity issues. The best LSTM model uses the Tanh activation function with 30 LSTM units, 10 timesteps, and a dropout of 0.01, resulting in a MAPE value of 5.323%. The best ARIMAX model obtained was the ARIMAX (0,1,1) model, with a MAPE value of 0.55% for the test data and 0.61% for the training data. The research results, indicate that the higher accuracy of ARIMAX reflects its suitability for linear data such as gold prices, but the accuracy of LSTM which is below 10% still performs well for more complex data patterns.Keywords: gold price; forecasting; LSTM; arimax.