Towards Agentic RAG with Deep Reasoning: A Survey of RAG-Reasoning Systems in LLMs

This survey synthesizes both strands under a unified reasoning-retrieval perspective. We first map how advanced reasoning optimizes each stage of RAG (Reasoning- Enhanced RAG). Then, we show how retrieved knowledge of different type supply missing premises and expand context for complex inference (RAG-Enhanced Reasoning). Finally, we spotlight emerging Synergized RAG-Reasoning frameworks, where (agentic) LLMs iteratively interleave search and reasoning to achieve state-of-the-art performance across knowledge-intensive benchmarks. We categorize methods, datasets, and open challenges, and outline research avenues toward deeper RAG-Reasoning systems that are more effective, multimodally-adaptive, trustworthy, and human-centric.

dynamic, iterative interplay where reasoning actively guides retrieval, and newly retrieved knowledge, in turn, continuously refines the reasoning process. This trend is further exemplified by recent ”Deep Research” products from OpenAI1, Gemini2, Perplexity3, and others, which emphasize tightly coupled retrieval and reasoning (Zhang et al., 2025f). These systems employ agentic capabilities to orchestrate multi-step web search and leverage reasoning to comprehensively interpret retrieved content, solving problems demanding in-depth investigation.

Traditional RAG methods first retrieve relevant documents, then concatenate the retrieved knowledge with the original query to generate the final answer. These methods often fail to capture the deeper context or intricate relationships necessary for complex reasoning tasks. By integrating reasoning capabilities across Retrieval, Integration, and Generation stages of the RAG pipeline, the system can identify and fetch the most relevant information, reducing hallucinations and improving response accuracy.4