Abstract: Onboard GNSS kinematic orbit determination is a primary approach for real-time orbit determination of low Earth orbit satellites. While pseudorange-based positioning typically achieves meter-level accuracy and carrier phase-based methods can reach centimeter-level precision, real-time orbit determination often faces challenges such as data loss, insufficient satellite visibility, gross errors, and undetected cycle slips, leading to discontinuities, increased errors, or filter divergence. Although dynamic models offer high accuracy and strong robustness, they require detailed force modeling and substantial computational resources, limiting their application mainly to post-processed precise orbit determination. This study adopts a GNSS kinematic-based approach, integrating a simplified dynamic model with high short-term predictive accuracy through adaptive “loose” or “tight” combinations of dynamics and GNSS observations, enabling dynamic-assisted real-time precise orbit determination. In the real-time experiments, a 40×40 Earth gravity field model was used, along with CNT real-time orbit and clock products and BeiDou Satellite Navigation System (BDS)/GPS dual-system data. Results show meter-level accuracy with pseudorange, and centimeter-level accuracy with carrier phase observations. Notably, meter-level discontinuities caused by cycle slips were effectively mitigated through dynamic assistance, achieving consistent centimeter-level accuracy with root mean square (RMS) values of 8.2 cm (radial), 8.1 cm (along-track), and 5.0 cm (cross-track) over the entire observation period.