In e-commerce product review mining, Aspect-Based Sentiment Analysis (ABSA) aims to extract fine-grained opinions toward specific aspects in reviews. Existing ABSA benchmarks have expanded from aspect-level sentiment classification to opinion-aware tuple extraction. However, they still provide limited support for evaluating implicit aspects with fine-grained specificity. In particular, implicit aspects are often omitted, collapsed into placeholders, or mapped to predefined categories, which can obscure what reviewers actually evaluate. We introduce OASIS, a review-level Open-vocabulary benchmark for Aspect-Opinion-Sentiment (AOS) extraction with Implicit-aspect Specificity. OASIS contains 6,775 Amazon reviews across eight product domains and 21,054 annotated tuples, where implicit aspects are directly represented as contextually inferred noun phrases rather than closed-set labels. This design yields an implicit-aspect-heavy evaluation setting, with 70.10% of tuples involving implicit aspects. We evaluate a range of open-source and proprietary LLMs under fine-tuned and few-shot settings, showing that full AOS extraction remains challenging even for strong models. We further show that open-vocabulary aspect phrases in OASIS can provide task-useful information. Our benchmark and experimental code is available at: https://github.com/DILAB-HYU/OASIS.