A 3,266-question multiple-choice benchmark covering the full breadth of the
wine domain — from viticulture and winemaking to wine regions, grape
varieties, producers, and the wine business. Designed to evaluate the
factual recall, comparative reasoning, and applied-decision capabilities
of large language models against expert-vetted, source-anchored knowledge.
Paper / repo: Anonymized for double-blind review
Track: NeurIPS 2026 Evaluations & Datasets (E&D)… See the full description on the dataset page:
https://huggingface.co/datasets/oenobench-anon/oenobench.
