Inductively coupled plasma (ICP) generators involve complex interactions between electromagnetic, thermal, and chemical processes, which makes direct diagnostics difficult. To clarify these coupling mechanisms, a two-dimensional axisymmetric model of an argon ICP torch operating at kilopascal pressure is developed using COMSOL Multiphysics under local thermodynamic equilibrium (LTE) and non-equilibrium (NLTE) assumptions. A two-dimensional axisymmetric magnetohydrodynamic (MHD) model is established, which combines electromagnetic induction, convective-radiative heat transfer, and a seven-reaction argon plasma chemistry mechanism. The LTE model assumes that the temperature of all species is uniform, while the NLTE model independently solves for the electron temperature (T_e) and gas temperature (T_g), thereby accounting for incomplete energy exchange between electrons and heavy particles. At a discharge power of 1000 W and a working pressure of 10 kPa, the LTE model predicts a peak temperature of approximately 8200 K, concentrated around the induction coils. In contrast, the NLTE model yields a maximum gas temperature of about 5990 K, with the hot zone shifted downstream. The NLTE model reveals a clear two-temperature structure: T_e peaks near the coil wall (~0.93 eV), while T_g reaches its maximum downstream, indicating a pronounced thermal non-equilibrium state where electrons are preferentially heated by the induced field. The calculated skin depth (~11.3 mm) coincides with the region of strongest electromagnetic energy deposition. Species analysis shows that the plasma core is dominated by ground-state argon (Ar) (>99%), while excited argon (Ar*) and argon ions (Ar⁺) increase notably near the coil region, confirming that excitation and ionization processes are localized within the skin layer. Furthermore, comparison between the 5 kPa and 10 kPa cases shows that as pressure decreases, the difference between T_e and T_g increases, indicating enhanced thermal non-equilibrium due to reduced collisional coupling. Overall, the results highlight that LTE and NLTE assumptions lead to markedly different predictions of temperature and energy coupling at kilopascal pressures. The NLTE model more realistically captures delayed energy transfer and spatial temperature decoupling, offering new insights into the electromagnetic-thermal-flow interactions of ICP discharges and providing a modeling reference for designing ICP-based high-enthalpy plasma wind tunnel and realizing related aerospace applications.