VariantScape-variant-extraction is an English biomedical literature dataset derived from the VariantScape literature-mining pipeline for precision oncology. The dataset contains article-level records retrieved from OpenAlex and curated for downstream natural-language-processing and large-language-model-based biomedical entity extraction. The version documented here contains 797 English biomedical publication records from 2014–2024, including article identifiers, titles, abstracts where available, publication metadata, human-curated variant annotations, and model-generated variant-extraction outputs from several large language models. The dataset is intended to support research on genetic variant recognition, structured biomedical information extraction, model-output evaluation, and literature mining for precision oncology. Because the records include source article titles and abstracts, reuse of those text fields remains subject to the rights and licenses of the original publications; the recommended license below applies only to the authors' annotations, curation metadata, and derived model-output fields.