This work presents BERTbek, one of the first monolingual transformerbased language models specifically for the Uzbek language. BERTbek is trained on a morphologically-sensitive tokenizer, utilising various text sources and training corpus sizes to evaluate the performance from different perspectives. Three variants of BERTbek — BERTbek-Wiki, BERTbek-News-Small, and BERTbek-News-Big — are trained and evaluated on various NLP tasks including Sentiment Analysis, Text Classification, and Named Entity Recognition (NER). Evaluation results demonstrate BERTbek-News-Big outperforming multilingual baselines including mBERT and other available models on prominent NLP tasks.