Karst-related carbonate fracture-cavity reservoirs play a vital role in global oil and gas field development. Especially under deep to ultra-deep conditions, their internal structures and filling-modification processes exhibit extreme complexity. Identifying the types and degree of fillings in paleokarst caves carries significant theoretical and practical value for evaluating effective reservoir space, optimizing development strategies, and tapping remaining oil potential. Based on an extensive review of the literature, this study proposes a systematic classification scheme for the filling phases and detrital filling phases of karst caves, highlighting key advancements in the geological understanding of internal cave filling structures. The article summarizes the current models of karst cave filling in the Tahe Area, focusing on technological progress in identifying and predicting filling materials and determining the degree of filling in paleokarst caves. Progress in identifying cave filling facies is primarily reflected in the genetic classification of modern surface cave detrital filling facies and the categorization of paleokarst cave fillings. Early methods for identifying and predicting cave filling materials and assessing filling degrees relied on qualitative and semi-quantitative approaches using logging and seismic data. With the advent of artificial intelligence (AI) technology, the application of machine learning’s powerful generalization capabilities to identify and predict filling materials and degrees has emerged as a cutting-edge research direction in this field. The classification of filling modes in paleokarst caves suggests utilizing the coupling relationship between hydrogeology and cave development within the hierarchical structure framework of the paleokarst fracture-cave system. This approach, combined with the types of internal filling materials revealed by actual drilling data, facilitates the construction of filling models. However, current classifications of filling types in paleokarst caves primarily focus on differences in rock physical components, without adequately reflecting the dynamic mechanisms of filling formation. Additionally, the accuracy of identifying cave fillings remains insufficient, hindering the comprehensive determination of the sequence of fillings within caves. Currently, seismic inversion technology, commonly used for predicting cave fillings, can only estimate mud content and fails to accurately evaluate the degree of filling for all materials. Consequently, predicting the spatial distribution of filling degrees in paleokarst underground river networks requires further research and development. In light of these challenges, this article argues that leveraging AI technology to identify and predict the types and degrees of cave filling materials represents a promising trend. Future research should focus on improving the representativeness of sample sets, as well as the accuracy and generalization capabilities of prediction networks.