Named entity recognition (NER) is one of the fundamental tasks in biomedical text mining, aiming to extract specific types of biomedical entities from texts. Compared to general-domain texts, biomedical texts often contain nested entities and local dependencies among entities, which pose challenges to existing NER models. To address these issues, we propose a nested biomedical NER method based on RoBERTa and Global Pointer (RoBGP). First, we dynamically obtain word-level semantic representations using the RoBERTa-wwm-ext-large model. Second, A BiLSTM model is used to capture long-distance semantic information in a bidirectional manner. Finally, we utilize a global pointer network to recognize all nested entities from innermost to outermost. Experimental results on the Chinese medical dataset CMeEE show that our method achieves the F1 score of 67.69%, outperforming other advanced models by 0.32% to 5.11%, validating the effectiveness of our approach in biomedical NER.