Curtis Alexander from NOAA/ESRL gave a webinar talk titled "Progress and plans for the operational HRRR model." The talk can be found here.
Below is just one slide from his presentation. It shows forecast skill of different methods of forecasting the weather. A persistence forecast, saying the weather will be the same later as it is now, has no skill beyond an hour. Extrapolation, or saying the weather trend will continue, does pretty good for an hour, but skill quickly decays. The HRRR model (dark green line) produce a better forecast than extrapolation after the first hour. After the first forecast hour the HRRR model is more accurate than persistence and extrapolation. As with all weather models, the forecast skill of the HRRR degrades over time, but still there is more skill in a HRRR forecast at six hours than a simple persistence or extrapolated forecast.