Agentic-R: Learning to Retrieve for Agentic Search
Paper: Agentic-R: Learning to Retrieve for Agentic Search (arXiv, 2026-01-17) Authors: Wenhan Liu, Xinyu Ma, Yutao Zhu, Yuchen Li, Daiting Shi, Dawei Yin, et al. (7 authors) Code: https://github.com/8421BCD/Agentic-R
Key points
- Problem: Agentic search has emerged as a powerful paradigm in which an agent interleaves multi-step reasoning with on-demand retrieval to solve complex questions. However, how to design a retriever specifically for agentic search remains largely underexplored. Existing search agents typically rely on similarity-based retrievers, and similar passages are not always useful for final answer generation.
- Core idea: Unlike retrievers designed for single-turn retrieval-augmented generation (RAG) that rely only on local passage utility, Agentic-R measures passage utility in multi-turn agentic search using both local query-passage relevance and global answer correctness.
- Iterative training strategy: The search agent and the retriever are optimized bidirectionally and iteratively. Instead of training a retriever once with fixed questions (as in RAG), Agentic-R is continuously improved using evolving, higher-quality queries generated by the agent.
- Results: Extensive experiments on seven single-hop and multi-hop QA benchmarks demonstrate that Agentic-R consistently outperforms strong baselines across different search agents.
- Specific quantitative results should be verified against the original PDF tables.
- Related reading: surveys and systems on RAG + reasoning, bootstrapped search agents (e.g., AceSearcher), and agentic information retrieval.
- Original paper: Agentic-R: Learning to Retrieve for Agentic Search, arXiv:2601.11888, https://arxiv.org/abs/2601.11888
Abstract (original)
> Agentic search has recently emerged as a powerful paradigm, where an agent interleaves multi-step reasoning with on-demand retrieval to solve complex questions. Despite its success, how to design a retriever for agentic search remains largely underexplored. Existing search agents typically rely on similarity-based retrievers, while similar passages are not always useful for final answer generation. In this paper, we propose a novel retriever training framework tailored for agentic search. Unlike retrievers designed for single-turn retrieval-augmented generation (RAG) that only rely on local passage utility, we propose to use both local query-passage relevance and global answer correctness to measure passage utility in a multi-turn agentic search. We further introduce an iterative training strategy, where the search agent and the retriever are optimized bidirectionally and iteratively. Different from RAG retrievers that are only trained once with fixed questions, our retriever is continuously improved using evolving and higher-quality queries from the agent. Extensive experiments on seven single-hop and multi-hop QA benchmarks demonstrate that our retriever, termed \ours{}, consistently outperforms strong baselines across different search agents. Our codes are available at: https://github.com/8421BCD/Agentic-R.
Context and significance
The work sits at the intersection of agentic search and large-scale retrieval systems. In the classic search stack (recall → ranking → generation), the LLM era introduces new variables: reasoning budget and action space (whether to retrieve, how many times, which tools to call). Agentic-R treats the retriever itself as a learnable, co-evolving component of the agent loop rather than a fixed similarity-based module.
Broader implications for search/retrieval practitioners:
1. Architecture: Cascaded retrieve-rerank-generate remains mainstream, but the agentic paradigm makes retrieval count and strategy themselves learnable. 2. Training signal: Global answer correctness (outcome-level reward) can complement local relevance signals when training retrievers for multi-turn settings. 3. Co-optimization: Joint, iterative optimization of agent and retriever yields higher-quality queries over time. 4. Evaluation: Task success and multi-hop reasoning quality matter beyond static nDCG-style relevance metrics.