A new benchmark called TSQueryBench has been developed to evaluate the performance of large language models (LLMs) in generating explanations for time series data. The benchmark consists of 500 synthetic time series instances paired with correct, partially correct, and incorrect explanations. Six LLMs were evaluated across four tasks: explanation generation, relative ranking, independent scoring, and multi-anomaly detection. The results show that while LLMs can generate accurate explanations, they are more reliable at evaluating the correctness of others' explanations than generating their own. This suggests that using LLMs as judges for time series explanations is a scalable approach. --- Why it matters: These findings matter to researchers in AI because they provide insight into the strengths and weaknesses of large language models in complex tasks like time series analysis, which has significant implications for applications such as finance and healthcare. Source: https://arxiv.org/abs/2604.02118