With the rapid development of data-driven meteorological models, the use of neural networks for nowcasting has shown great potential. Therefore, evaluating the capabilities of existing large-scale meteorological models is crucial for improving their performance in nowcasting. In this paper, we propose LWMEval, a benchmark tool specifically designed to assess the performance of large-scale meteorological neural network models in six-hour nowcasting. Our evaluation includes seven large-scale neural network models, covering five key nowcasting tasks. By employing a multidimensional metric system and seasonal analysis, we comprehensively examine the models’ nowcasting capabilities and computational efficiency. Existing evaluations often focus on the performance of neural network models in medium-term weather forecasting, with a lack of dedicated benchmarks for nowcasting. This benchmark tool aims to fill that gap, providing valuable insights for enhancing the accuracy and efficiency of large-scale weather models in six-hour nowcasting.