Abstract: Automatic text summarization has undergone transformative advancements into the current era of large language models(LLM). In recent times, numerous domestic open-source large models, having intensified training on Chinese corpora, demonstrate remarkable progress. To assess the performance of Chinese open-source large language models on the task of abstractive text summarization, this paper selects ChatGLM2-6B, Baichuan2-7B, and InternLM-7B as subjects. These models are tested on two Chinese summarization datasets using various prompts for both zero-shot and few-shot summarization. The results show that ChatGLM2-6B and Baichuan2-7B typically generate coherent, detailed summaries via zero-shot, albeit lacking conciseness.On the other hand, few-shot method allows LLMs to generate brief summaries, but results in the omission of many key information.Additionally, the models exhibited issues such as repetition loops, hallucinations, and factual inconsistency.