Reasoning-enhanced Retrieval-Augmented Generation

This comparison evaluates 15 Reasoning-Enhanced RAG models, which integrate reasoning capabilities across the Retrieval, Integration, and Generation stages of the RAG pipeline. These models are categorized into three main groups: Retrieval Optimization: Focused on improving the relevance and quality of retrieved results. Integration Enhancement: Designed to assess the relevance of retrieved information and effectively merge heterogeneous evidence, thereby preventing irrelevant content from disrupting generation. Generation Enhancement: Aims to reduce the generation of unfaithful or factually incorrect content, improving overall output fidelity.