The mainstream methods of timeline summarization (TLS) task generally follow the four-stage pipeline, including event detection, event clustering, cluster summarization, and timeline construction. Currently, large language models (LLMs) have demonstrated significant advantages in text generation and semantic understanding, but their application in TLS remains under-explored. To our knowledge, this study pioneers the application of LLMs for comprehensive event extraction from individual news articles. In terms of event clustering, facing complicated linguistic scenarios, high-performing LLMs’ deployment costs are prohibitively high and they are subject to external resources. Conversely, some lightweight LLMs are open-sourced and can be deployed locally. Therefore, this study proposes a hybrid approach that combines coarse-grained clustering with fine-grained clustering. Through Llama2-13B, the model gains the capability to distinguish identical events in complex scenarios, thus can improve the effectiveness of clustering process. Subsequently, the cluster summarization is generated using LLM. The timeline is derived based on two attributes: cluster count and relevant count. Experimental results show the efficacy of our method and our results showed the state-of-the-art performance on three open-sourced datasets.