We model future EUR/USD exchange rate realized volatility (RV) within a class of heterogeneous autoregressive (HAR) models augmented by implied volatilities (IVs). The existing literature has almost unanimously employed IVs from options with one-month maturities; however, our in-sample analysis shows that using IVs from options with a shorter maturity (of one day and one week) might be more relevant when explaining the volatility of the next day and week. In general, IVs are more useful in predicting future RV than past RVs (daily, weekly and monthly averages). At the same time, RVs seem to contain only small incremental predictive power compared to IVs. The out-of-sample results strengthen our in-sample results, as they show the increased predictive power of the models with implied volatility up to 17.3% for one-day-ahead, 42.1% for one-week-ahead, and 22.8% for one-month-ahead forecasts. Additionally, the superior set of models contains only volatility model specifications with IVs. Our results hold not only for individual forecast models but also for combinations of volatility forecasts. We show that increased forecasting accuracy is stable across time and that it is achieved during periods of high market volatility. Our study also provides new evidence that implied volatility from short-lived options as a serious contender for modeling realized volatility