This paper evaluates the real-time forecast performance of alternative Bayesian autoregressive (AR) and vector autoregressive (VAR) models for the Australian macroeconomy. To this end, we construct an updated vintage database and compare the predictive ability of a wide set of specifications that takes into account almost all possible combinations of nonstandard errors existing in the current literature. In general, we find that the models with flexible covariance structures can improve the forecast accuracy as compared with the standard variant. For forecasting GDP, both point and density forecasts consistently suggest small VARs tend to outperform their counterparts while AR models often predict inflation better. With the unemployment rate, large VAR models provide superior forecasts to the alternatives at almost all forecast horizons. The forecasting performance of these models slightly changes when we consider the first, second, and latest-available vintage as actual values, highlighting the importance of using real-time data vintages in forecasting.