Open-source software (OSS) has revolutionized modern software development by fostering collaboration across diverse teams. However, as OSS projects grow in size and complexity, managing license compliance has become increasingly challenging. A critical issue lies in accurately recognizing and interpreting the varied clauses within OSS licenses, particularly when multiple licenses coexist, each with distinct permissions, obligations, and restrictions. Traditional license analysis tools, often rule-based, struggle to identify nuanced conflicts between license clauses, leading to potential compliance risks. In response to these challenges, this paper presents a fine-grained, high-quality dataset of 634 SPDX-certified licenses, annotated with 3,396 individual clauses across 14 categories. Each clause has been meticulously reviewed and validated using model-assisted checks to ensure accuracy, providing a solid foundation for detailed clause-level analysis. To improve clause recognition and conflict detection, we introduce ClauseBench, a benchmarking framework that leverages large language models (LLMs) to detect and interpret license clauses with high precision. ClauseBench improves detection accuracy by 50 % compared to traditional document-level methods and significantly reduces hallucination rates by focusing on individual clauses, where precise distinctions in legal language are crucial. Additionally, we implemented a contextual prompt engineering strategy to optimize model performance, achieving 90% accuracy in clause identification. Our framework sets a new standard for automated license conflict detection in OSS, demonstrating the potential of LLMs to manage the complexities of legal text interpretation. This work not only advances the field of license analysis but also opens the door to future research on integrating LLMs with OSS compliance tools.