Verified LLM-Generated Scientific References: 146,001 resolved references from 9 large language models

Nine large language models (GPT-4o, Claude 3.5 Sonnet, DeepSeek-R1, DeepSeek-V3, Gemini 1.5 Flash/Pro, Gemini 2.0 Pro, Llama 3.1-405B, and Llama 3.3-70B) were prompted to produce literature reviews with references for each of 497 OpenAlex topics spanning 26 fields, with 10 repetitions per model per topic. This dataset contains 146,001 generated reference instances that were verified against Crossref, OpenAlex, Semantic Scholar, and arXiv, and resolved to unique, real OpenAlex journal articles, representing 54,457 distinct works. Each row pairs the model’s raw output with the corresponding ground-truth record and includes bibliometric attributes of that record. The files are currently restricted pending publication of the associated manuscript, and access requests are reviewed by the authors. See README.md for the full data dictionary, sampling frame, and known issues.