We retrieve sea levels in polar regions via GNSS reflectometry (GNSS-R), using signal-to-noise ratio (SNR) observations from eight POLENET GNSS stations. Although geodetic-quality antennas are designed to boost the direct reception from GNSS satellites and to suppress indirect reflections from natural surfaces, the latter can still be used to estimate the sea level in a stable terrestrial reference frame. Here, typical GNSS-R retrieval methodology is improved in two ways, 1) constraining phase-shifts to yield more precise reflector heights and 2) employing an extended dynamic filter to account for the second-order height rate of change (vertical acceleration). We validate retrievals over a 4-year period at Palmer Station (Antarctica), where there is a co-located tide gauge (TG). Because ice contaminates the long-period tidal constituents, we focus on the main tidal species (daily and subdaily), by employing a deseasonalization filter. The difference between sub-hourly GNSS-R retrievals of the ocean surface and TG records has a root-mean-square error (RMSE) of 15.4 cm and a correlation of 0.903, while the tidal prediction has a RMSE of 1.9 cm and a correlation of 0.998. There is excellent millimetric agreement between the two sensors for most eight major tidal constituents, with the exception of luni-solar diurnal (K1), principal solar (S2), and luni-solar semidiurnal (K2) components, which are biased in GNSS-R due to the leakage of the GPS orbital period. We also compare the GNSS-R tidal constituents from seven additional POLENET sites, without co-located TG, to global and local ocean tide models. We find that the root-sum-square-error (RSSE) of eight major constituents varies between 26.0 cm and 56.9 cm for different models. Given that the agreement in tidal constituents between the TG and GNSS-R was better at Palmer Station, we conclude that assimilating the GNSS-R retrievals into tidal models would improve their accuracy in Antarctica and Greenland, provided that care is exercised to avoid the orbital period overtones and also sea ice.