Satellite positioning is an essential part of modern society. The accuracy of satellite positioning is constantly getting better and correspondingly the accuracy of the geoid model, used for the conversion of measured heights to the heights in the national height system, should be improved. The accuracy of the geoid model is in a crucial role when considering a future height reference system based on GNSS and a geoid model. Currently, the general consensus is that the geoid model should have an accuracy of 1 cm and better. In this study we investigated what the accuracy and density of the input gravity dataset should be in Finland to reach geoid model accuracies such as 1 cm or better. In the project we looked at the influence of the different input data on the total geoid model error. The study looked at the omission and commission errors for two areas in Finland, a rather flat area in the South of the country and a rougher area in the North. The project results show that the 1.4 cm accuracy of the recently issued FIN2023N2000 model, based on terrestrial gravity data with a resolution of 5 km and an accuracy up to 2 mGal, is close to the best we can obtain with the current gravity data in Finland. To obtain a geoid model accuracy below 1 cm in Finland, gravity data is needed with a resolution of 3 km or better and uncertainties of 1 mGal or better, that are homogenous and normally distributed. This implies that a gravity dataset twice as dense as the current dataset is needed in the future.