A systematic comparison of contextualized word embeddings for lexical semantic change is presented. The study evaluates state-of-the-art models for Graded Change Detection (GCD) under equal conditions across eight languages. It breaks down the LSC problem into Word-in-Context (WiC) and Word Sense Induction (WSI) tasks, comparing models across these levels. The results show that APD outperforms other approaches for GCD, while XL-LEXEME excels in WiC, WSI, and GCD, performing comparably to GPT-4. The study highlights the need to focus on how, when, and why word meanings change, rather than just the extent of semantic change. It also compares GPT-4 with contextualized models, finding that XL-LEXEME is a more affordable and effective alternative. The evaluation reveals that form-based approaches significantly outperform sense-based ones, with APD being the best for GCD. The study emphasizes the importance of evaluating both (i) and (ii) in the LSC framework, as they are equally important as (iii). The results indicate that XL-LEXEME is a strong candidate for LSC tasks, offering a scalable and affordable solution. The study also notes the limitations of using GPT-4 due to cost and accessibility, and highlights the need for further research into sense-based approaches. Overall, the study provides a comprehensive evaluation of contextualized word embeddings for LSC, establishing clear settings for future comparisons and evaluations.A systematic comparison of contextualized word embeddings for lexical semantic change is presented. The study evaluates state-of-the-art models for Graded Change Detection (GCD) under equal conditions across eight languages. It breaks down the LSC problem into Word-in-Context (WiC) and Word Sense Induction (WSI) tasks, comparing models across these levels. The results show that APD outperforms other approaches for GCD, while XL-LEXEME excels in WiC, WSI, and GCD, performing comparably to GPT-4. The study highlights the need to focus on how, when, and why word meanings change, rather than just the extent of semantic change. It also compares GPT-4 with contextualized models, finding that XL-LEXEME is a more affordable and effective alternative. The evaluation reveals that form-based approaches significantly outperform sense-based ones, with APD being the best for GCD. The study emphasizes the importance of evaluating both (i) and (ii) in the LSC framework, as they are equally important as (iii). The results indicate that XL-LEXEME is a strong candidate for LSC tasks, offering a scalable and affordable solution. The study also notes the limitations of using GPT-4 due to cost and accessibility, and highlights the need for further research into sense-based approaches. Overall, the study provides a comprehensive evaluation of contextualized word embeddings for LSC, establishing clear settings for future comparisons and evaluations.