Biomass is one of the most abundant renewable carbon resources, and pyrolysis is a key thermochemical route for converting macromolecular biomass into bio-oil, gases and solid char. This review synthesizes advances in kinetic modeling of biomass pyrolysis by tracing the transition from lumped thermogravimetry-based global descriptions through semi-detailed component schemes to extensive, mechanism-based, detailed reaction networks, emphasizing diagnostics, parameterization, and model validation. Global kinetic frameworks represent overall conversion with lumped reactions that correlate macroscopic behavior under defined heating protocols and provide empirical estimates of char, condensables, and permanent gas yields. Semi-detailed approaches subdivide these collective pathways into multiple competitive and consecutive steps that represent cellulose, hemicellulose, and lignin sub-mechanisms, and introduce intermediate entities such as activated cellulose to enable prediction of smaller lumped species and volatile compounds. Detailed mechanistic models construct reaction networks comprising hundreds to thousands of elementary steps and species to describe molecular-level routes from polymer depolymerization to the formation of radicals, low molecular weight oxygenates, aromatic intermediates, and heavy tar precursors. The development of these networks has been supported by high-resolution, time-resolved analytical platforms, including pyrolysis gas chromatography–mass spectrometry (Py-GC/MS) with flame ionization detection (FID), two-dimensional GC, and synchrotron vacuum ultraviolet photoionization mass spectrometry (SVUV-PIMS), together with reactor designs that resolve condensed and gas phases on millisecond-to-second timescales and yield quantitative constraints for elementary rates. Computational contributions include density functional theory (DFT) calculations to obtain bond dissociation energies and transition state barriers for critical steps, automated mechanism-generation software to assemble reaction families, and machine learning surrogates with data-driven parameter optimization to reduce simulation cost while retaining predictive fidelity. Representative models range from carbohydrate-focused schemes with about one hundred elementary steps to hemicellulose mechanisms with several hundred reactions and lignin networks containing thousands. These models reveal cooperative depolymerization in carbohydrates and radical-driven fragmentation in lignin. Current efforts couple condensed-phase and gas-phase processes to describe secondary cracking and rearrangement reactions that shape tar and oxygenate formation. Sensitivity and uncertainty analyses identify key reaction families and kinetic parameters that control yields of specific products, supporting targeted experiments for model refinement. Modular representations treat lignin as an assembly of structural units and describe feedstock variability through monomer ratios, linkage patterns, and molecular weight, helping connect laboratory mechanisms with industrial materials. Validation studies reproduce main gas and char trends but also show strong dependence on cellulose crystallinity, particle size and other feedstock properties, underscoring the need for standardized feedstock models and integrated, multi-scale kinetic frameworks. Overall, the continuous evolution from empirical to mechanistic and detailed kinetic models has greatly deepened the understanding of biomass pyrolysis chemistry. Future work should focus on multi-scale model integration, heterogeneous catalytic pyrolysis, and data-driven kinetic parameter optimization to achieve predictive, process-oriented modeling for industrial applications.