Background: Genome-Scale Metabolic Models (GSMMs) are comprehensive mathematical representations of a cell’s biochemical reactions and associated genes and metabolites. They offer a structured framework to study metabolism at a systems biology level. GSMMs are usually built from generic models like Human1, which cover all described human metabolism, and integrate multi-omics data (i.e., transcriptomics, metabolomics, proteomics) using algorithms such as CORDA and GIM3E to generate context-specific models that reflect the metabolic activity of specific cell types or conditions. Metabolic behavior is then simulated using Flux Balance Analysis (FBA). However, while this top-down approach captures the complexity of cellular metabolism, it often results in biologically unrealistic flux distributions due to the limited coverage of experimental metabolomic and proteomic data. To overcome this limitation, we propose a complementary bottom-up strategy using reduced models of central metabolism (approximately 400 reactions). These simplified, well-characterized, and experimentally validated models serve as a controlled and reliable core to guide flux distributions in full-scale GSMMs, leading to more realistic simulations. Moreover, besides serving as a structural core, these reduced models help explore aspects like the trade-off between model specificity and predictive performance based on environmental constraint flexibility. We apply this integrative methodology to study tumor metabolic reprogramming in the colorectal cancer microenvironment, using extensive multi-omics datasets and in vitro models developed by our research group. These combined top-down and bottom-up strategies enable the construction of more reliable and biologically relevant context-specific GSMMs with high predictive power, enhancing their utility in identifying metabolic vulnerabilities and designing effective drug combinations.