Global Navigation Satellite System reflection (GNSS-R) signals can be leveraged to measure reflecting surface height. However, the code-based altimetric retrievals present biases that vary across reflection tracks, whose origin and nature remain unclear. This study utilizes Spire Global Inc.’s delay Doppler maps (DDMs) of GPS L1CA, GAL E1C, and BDS B1C reflection signals over sea ice to characterize the variation of biases and the precision of height retrievals. Employing data pairs of the same reflection signals collected by different antennas on a receiver, we observe that the height retrievals agree well to each other with similar biases from the mean sea surface (MSS) model. Height biases of the same transmitter and receiver also vary across acquisition time. Moreover, based on abundant reflection data over Hudson Bay, the height biases of different combinations of GNSS system and receiver follow normal distributions. After examining bias magnitude versus SNR, smaller biases can be observed at higher SNR levels. As for precision, it is generally on the order of meter to submeter for all GPS, GAL, and BDS height measurements, and improves with SNR, demonstrated by the same retrievals over Hudson Bay. GAL and BDS retrievals have similar precision from ~ 1.0 m at a SNR of 3 dB to ~ 0.5 m at 13 dB, better than that of GPS from ~ 1.5 m to ~ 0.5 m over a similar SNR range. The better precision of GAL and BDS retrievals can be attributed to narrower waveforms and finer DDM delay resolutions. A root mean squared difference (RMSD) of 0.83 m can be achieved between the averaged bias-corrected height observations of all GNSS systems and receivers and the MSS over Hudson Bay. Our study contributes to understanding the randomly varying height biases and designing instruments to facilitate accurate and precise altimetric observations.