This study investigates the relationship between the number of tokens in Uzbek-language texts and the percentage share of unique tokens. The purpose is to determine the form of this relationship under a linear model and a Heaps’-law-type power-law model and to compare their descriptive performance. The scientific significance of the study lies in providing empirically grounded evidence on this regularity specifically for the Uzbek language, where it has not previously been examined in this form. The practical motivation arises from the author’s broader dissertation research on text similarity, text summarisation, text classification, and text clustering using TF-IDF, whose vector dimensionality is determined by vocabulary size and therefore affects memory requirements. The study analysed a corpus of 132 Uzbek-language texts of different sizes, calculated token counts and unique-token shares, and modelled their relationship using linear and Heaps’-law-type power-law regression. The results show that Heaps’ law describes the relationship considerably more accurately than the linear model, with goodness of fit of approximately 94% compared with 69%, respectively. The findings demonstrate that unique-token growth is sublinear rather than proportional to text length. This provides a basis for forecasting vocabulary size, vector dimensionality, and memory requirements of TF-IDF-based applications in advance, helping to anticipate and mitigate the curse of dimensionality in Uzbeklanguage text processing.