Abstract: Direct air capture (DAC) is an important negative-emission route for achieving carbon neutrality, and adsorption-based DAC has become the mainstream approach owing to the high CO₂ selectivity and good engineering adaptability of solid sorbents. However, capturing CO₂ from ambient air with an extremely low CO₂ concentration and dynamically varying temperature and humidity faces the dual challenges of high energy consumption and high cost, whose resolution relies on multiscale modeling spanning microscopic mechanisms to macroscopic systems. Taking the stepwise transfer of parameters as the main thread, the progress of multiscale modeling of adsorption-based DAC from molecular simulation to system integration is reviewed, covering the molecular, reactor and system scales and their coupling. The molecular scale is the starting point of this chain, where grand canonical Monte Carlo and molecular dynamics simulations predict intrinsic properties such as equilibrium uptake, competitive selectivity and diffusivity, and reveal the competitive and cooperative adsorption mechanisms of CO₂ and H₂O at the molecular level: moderate humidity promotes CO₂ capture by participating in the reaction pathway and improving mass transfer, whereas excessive water suppresses CO₂ diffusion by occupying sites and blocking pores. These intrinsic parameters are then transferred upward, correlated through adsorption isotherm and kinetic models and coupled into the mass, energy and momentum conservation equations of the fixed bed, so that the dynamic breakthrough process is simulated at the reactor scale. The resulting breakthrough behavior and cyclic performance are further transferred to the system scale, where process simulation, energy decomposition and techno-economic analysis of temperature vacuum swing adsorption (TVSA) and steam-assisted temperature vacuum swing adsorption (S-TVSA) cycles convert process parameters into specific energy consumption and the levelized cost of capture. Existing analyses show that the capture cost is highly sensitive to sorbent performance, cycle operating parameters and energy price, and remains far higher than that of post-combustion capture, with thermal energy constituting the main part of the energy demand, highlighting the importance of water management and the coordinated optimization of materials and processes. Across the scales, the stepwise transfer of intrinsic parameters connects the molecular, reactor and system levels, which is the basis distinguishing multiscale modeling from single-scale analysis. Nevertheless, existing models are mostly based on idealized conditions and can hardly describe the effects of dynamic temperature and humidity fluctuations in real air, while coupling with intermittent renewable energy imposes higher requirements on the dynamic response of models; moreover, high-fidelity models are computationally expensive and difficult to embed directly into iterative process optimization. Machine learning offers a new route by constructing cross-scale surrogate models from data generated by physical models, alleviating data scarcity while reducing computational cost. In the future, deepening the integration of physics-based and data-driven models and developing intelligent modeling methods are expected to drive adsorption-based DAC toward low-cost and large-scale applications.