This paper presents an investigation of the precise point positioning (PPP) performance of a combined solution from BDS-2 and BDS-3 satellites. To simultaneously process different BDS signal observations, i.e., B1/B1C, B2/B2a and B3C, undifferenced and uncombined observations with ionosphere delay constrained by the deterministic plus stochastic ionosphere model are used in the basic model. Special attention is paid to code bias and receiver clock parameters in the derivation of the observation model. The analysis is carried out using more than one-month data for BDS-2 and BDS-3 collected at the CANB, DWIN, KNDY and PETH stations in the Asia-Pacific region. The results suggest that compared with BDS-2 alone, the BDS-2 and BDS-3 solution provides significantly more accurate PPP, with increases of 28%, 21% and 5% in the up, north and east directions, respectively. In addition, the average root mean square error decreases to 0·21, 0·13 and 0·16 m for the three directions. Furthermore, the PPP convergence time for BDS-2 and BDS-3 is about 1·5 h and less than 1 h for the horizontal and vertical components, respectively, whereas that for BDS-2 alone is about 2·3 h for both directions.