Gold today maintains its critical role both in hedging activities and in industry. Being one of the important indicators of the market situation and the fact that the XAU/USD ounce price is used in pricing many financial instruments reveals the importance of gold price estimation. This study aims to contribute to the literature by proposing a deep learning-hyperparameter optimization method that can provide promising results in daily gold price prediction studies. Additionally, this study determines which input sequence length is more informative for gold price prediction for each model. For this purpose, this study uses the last 7-year XAU/USD ounce price and 10 features that may be related to gold, and predicts the next day’s XAU/USD ounce price with Long-Short Term Memory (LSTM), Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN), Temporal Convolutional Network, Recurrent Neural Network (RNN) deep learning methods. This research trains prediction models with both default parameters and Bayesian, Genetic algorithm and Grey-Wolf hyperparameter optimization methods for 8, 16, 32 and 64 window sizes. The prediction performance of the models is compared by Mean Squared Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Percentage Error (MAPE) and coefficient of determination (R2). Accordingly, this paper reveals that the GRU-Bayesian model shows the highest performance for window sizes of 16 and 32. Also, this study shows that Bayesian optimization performs better among hyperparameter optimizations.