WAM, SWAN and WAVEWATCH III® were implemented to the Finnish archipelago with a 0.1 nmi grid. A comparison with coastal wave buoy observations showed that the models agreed on the significant wave height, with biases and root-mean-square-errors (RMSE) differing at most 0.06 m. In a general sense, WAM propagated most long wave energy into the archipelago, while SWAN generated the highest local waves. The performance of WAVEWATCH III was wind direction dependent. The model tendencies caused them to disagree on the peak period near the coast, with differences in mean values being up to 1.4 s. The large scatter (RMSE>2 s) inside the archipelago was mostly explained by the ill-defined nature of the parameter in more complex wave conditions. The mean period had less scatter (RMSE<1.5 s), but changes in the upper integration frequency from 0.6 Hz to 1 Hz affected the bias by roughly 1 s in all models. WAM and WAVEWATCH III underestimated the high-frequency wave energy for certain wind directions, possibly because of a too small friction velocity. A wind forcing taken every 3 h from a 7.4 km operational atmospheric model was found to be sufficient to force the high-resolution wave models.