Abstract: With the accelerated development of new domains and capabilities, the design of novel reactors has become increasingly diverse. Constrained by application environments, compact core structures are required for space reactors and underwater unmanned vehicles (UUV). Correspondingly, the heterogeneity of external control rods and fuel assemblies poses challenges to traditional reactor physics calculation methods. In the context of the two-step deterministic calculation scheme, the accuracy of core neutronics calculations is largely dictated by the precision of homogenized few-group cross-sections. This study investigates complex assembly cross-section generation methods based on the advanced reactor neutronics analysis program system, SARAX. The SARAX system consists of the cross-section generation module TULIP and the core calculation module LAVENDER. Building upon the existing calculation algorithms of the cross-section generation code TULIP, a two-dimensional assembly fine modeling function was developed. Resonance self-shielding calculations were performed using the ultra-fine group method featuring a 1 968 group energy structure. Background cross-sections were calculated using the embedded self-shielding method (ESSM) coupled with a method of characteristics (MOC) fixed-source solver. Neutron transport calculations were performed using the MOC solver to obtain neutron flux distributions of various angular orders. Subsequently, spatial homogenization and energy group condensation were executed based on the 1 968 group neutron spectrum to generate few-group cross-sections for core calculations. A specific heat pipe reactor (HPR) scheme was selected as an example to validate the proposed method against continuous energy Monte Carlo reference solutions. This HPR scheme featured a square lattice arrangement of fuel pins and heat pipes, incorporating ex-core control systems. Since this reactor core assemblies could not be equivalently modeled as one-dimensional rings, two-dimensional assembly codes were required for detailed modeling. Numerical results demonstrate that the deviation in the assembly infinite multiplication factor (k_inf) calculated using the new few-group cross-sections remains within 200 pcm compared to the Monte Carlo code, while the deviation in the core effective multiplication factor (k_eff) is −243 pcm. Regarding the reactivity worth of the two control systems in the heat pipe reactor, the safety rod worth exhibits a −2.36% relative deviation from the reference solution, and the sliding reflector worth shows a −0.31% relative deviation. The computational accuracy demonstrates excellent agreement with the Monte Carlo code. Furthermore, over a 5 years reactor core life, the deviation in core k_eff remains within a stable range as burnup depth increases. Additionally, a comparative analysis of computational accuracy and efficiency is conducted between the ESSM and the improved Tone’s method. While maintaining computational accuracy, the resonance calculation efficiency of the few-group cross-section generation method based on the ESSM with ultra-fine group is enhanced by a factor of three compared to the improved Tone’s method.