Abstract: Based on the concept of dynamic zone partition, improved delayed detached eddy simulation (IDDES) modeling of high-Ma full-scale scramjets with more than 100 million cells was conducted for the integrated internal and external flow fields. A complete dynamic zonal combustion modeling framework was established, including dynamic zone flamelet model (DZFM), zonal dynamic adaptive chemistry (Z-DAC), and zonal in situ adaptive tabulation (Z-ISAT). The fidelity of the zonal modeling framework is preliminarily verified by the 115-million-cell modeling of a benchmark hypersonic combustor named REST, which was designed to operate at Mach 12. Through the idea of local flow-chemistry decoupling within each zone, DZFM not only accurately represents the local turbulence-chemistry interaction but also effectively improves the computational efficiency of turbulent combustion in the whole field. Z-DAC and Z-ISAT can further improve the resolving efficiency of chemical reactions in each zone by dynamically reducing the chemical mechanism and tabulating the thermochemical states. Then based on 125 and 140 million cells, respectively, the characteristics of hydrogen-fueled strut and pylon hypersonic combustors were comparatively analyzed for Mach 10. Both the pylon and strut structures induce obvious boundary layer separation and fore-body recirculation zone, resulting in long pre-combustion regions in front of the injection point in both combustors. Numerical analysis based on the Borghi diagram shows that the diffusion-dominated flame mode widely exists in the current hydrogen-fueled hypersonic combustor, and the bottleneck of efficiency improvement lies in efficient mixing. The pylon combustor has higher jet penetration depth and better near-field mixing, and thus the combustion efficiency of 80% is above the criterion of achieving net thrust. The specific impulse of 1234 s in the pylon combustor is also much higher than the 437 s in the strut combustor. Z-DAC reduces the computational cost of reaction systems in nearly half of the computational domain, especially in the fuel-free regions. Compared with the traditional finite-rate PaSR model, the DZFM model achieves an acceleration ratio of up to 11.