Large language models (LLMs) demonstrate impressive emergent reasoning capabilities, yet they remain constrained by fixed model parameters and finite context windows. Even when augmented with static retrieval, they can produce statements that are unsupported by, or even contradictory to, retrieved evidence. This leads to hallucinations, difficulty incorporating up-to-date information, and failures to correctly interpret or reconcile sources. Recent work addresses these limitations by tightly coupling step-by-step reasoning with interactive search, treating retrieval as an action within an iterative agentic loop rather than a one-shot preprocessing step. This study reviews frameworks that implement such reasoning-search loops by interleaving planning, querying, retrieval, integration, verification, and stopping. We analyze these systems through four architectural axes: retrieval coupling, controller design, knowledge source, and grounding strictness. We separately treat the learning signal as a cross-cutting dimension because prompting, supervision, preference optimization, and reinforcement learning can instantiate the same architecture while producing materially different search, stopping, and grounding behavior. We further distinguish benchmark difficulty from RSA-loop demand and synthesize controlled within-study ablations and quality-cost deltas, avoiding invalid pooling across heterogeneous model backbones, retrievers, corpora, and evaluation protocols. Our goal is to clarify the landscape of reasoning-search-augmented LLMs and outline a unified perspective on the design choices shaping future research.
Figures & tables
Fig. 1: Unified taxonomy of Reasoning–Search-Augmented LLMs. The top panel illustrates the iterative reasoning–search loop, while the bottom panel characterizes the design space along four orthogonal axes: retrieval coupling, controller design, knowledge source, and grounding strictness. Scheduled/Static retrieval is shown only as a reference baseline; learning signal is treated as a cross-cutting dimension.
Setting
Benchmark
Primary capability
Demand
Main limitation
Fact lookup
NQ; TriviaQA [ 52 , 53 ]
Open-domain retrieval and short-answer generation
L
Often solvable from one passage; static corpora and contamination limit evaluation of search control.
Cross-document evidence composition and supporting-fact reasoning
M
Fixed or provided contexts permit pre-retrieval or long-context inference; EM/F1 obscures retrieval control.
End-to-end RAG
FRAMES [ 57 ]
Factuality, retrieval, and multi-constraint reasoning
M
Small, Wikipedia-centered set that does not isolate controller behavior.
Verification
FEVER [ 58 ]
Evidence retrieval, claim support, and veracity classification
L
Static Wikipedia and synthetic claims allow fixed retrieval without adaptive search.
Noisy search
SEALQA [ 59 ]
Ambiguous, conflicting, time-sensitive, and distractor-heavy evidence
H
Small, version-sensitive set; LongSeal tests selection from supplied documents rather than active acquisition.
State-gated retrieval
SGR-Bench [ 60 ]
Configuring filters, hierarchies, and scopes to expose answer-bearing evidence
H
Only 100 tasks; preprint status and changing interfaces limit stability.
TABLE I: Representative benchmarks and their demand for an adaptive reasoning–search loop. L, M, and H denote low, moderate, and high RSA-loop demand; Var. denotes item-dependent demand.
Method
Coupling
Controller
Knowledge source / retriever
Ground.
Learn.
Panel A: Adjacent reference systems
RAG / FiD [ 4 , 21 ]
Sch.
Single / linear or fusion
Wikipedia; DPR
BE
SFT
Feedback-space planning [ 23 ]
Ad.
Search over candidate models
VAL and landmark feedback
N/A
P
GopherCite [ 49 ]
Sch.
Single / rerank
Open web; Google Search top-K
Hard
SFT+Pref
Panel B: Core RSA-LLMs
ReAct [ 9 ]
Ad.
Single / linear
Wikipedia API
BE
P
TABLE II: Representative systems across five design dimensions. Adjacent reference systems are excluded from cross-system counts.
Study
Controlled contrast
Type
kQ
Reported quality change
Efficiency change
Source
Panel A: Single-component and module-level evidence
TABLE III: Controlled quantitative evidence. Quality entries give median absolute change [min, max], positive-effect counts, and kQ benchmark–metric observations. S, C, M, and A denote single-component, conditional, multifactor, and aggregate-only evidence; studies are not pooled.
Fig. 2: Reported within-study deltas from Table III : (A) EM, accuracy, and success-rate; (B) F1 changes; and (C) resource reductions. Points denote individual results or medians across benchmarks; while whiskers show benchmark ranges, not confidence intervals. Marker style distinguishes single-component (S), conditional (C), and multifactor (M) contrasts. Results are not pooled across panels.
Fact-seeking question answering with large language models (LLMs) remains unreliable when answers depend on up-to-date or conflicting information. Although retrieval-augmented and tool-using LLMs reduce hallucinations, they often rely on implicit planning, leading to inefficient tool usage. We propose a modular framework that explicitly separates planning from factual retrieval and answer synthesis. A lightweight student planner is trained via a teacher-student framework to generate structured decompositions consisting of abstract reasoning steps and searchable fact requests. The supervision signals contain only planning traces and fact requests, without providing factual answers or retrieved evidence. At inference, the planner produces plans, while prompt-engineered modules perform retrieval and response synthesis. We evaluate the proposed framework on SEAL-0, an extremely challenging benchmark for search-augmented LLMs. Results show that supervised planning improves both accuracy and latency compared to monolithic reasoning models and prompt-based tool-augmented frameworks, demonstrating that explicitly learned planning structures are essential for reliable fact-seeking LLMs.
Auksarapak Kietkajornrit, Jad Tarifi, Nima Asgharbeygi
Large reasoning models such as DeepSeek-R1 and OpenAI o1 generate extended chains of thought spanning thousands of tokens, yet their integration with retrieval-augmented generation (RAG) remains fundamentally misaligned. Current RAG systems optimize for providing context before reasoning begins, while reasoning models require evidence injection during multi-step inference chains. We introduce ReaLM-Retrieve, a reasoning-aware retrieval framework that addresses this mismatch through three key innovations: (1) a step-level uncertainty detector that identifies knowledge gaps at reasoning-step granularity rather than token or sentence level; (2) a retrieval intervention policy that learns when external evidence maximally benefits ongoing reasoning; and (3) an efficiency-optimized integration mechanism that reduces per-retrieval overhead by 3.2x compared to naive integration. Experiments on MuSiQue, HotpotQA, and 2WikiMultiHopQA demonstrate that ReaLM-Retrieve achieves on average 10.1% absolute improvement in answer F1 over standard RAG (range: 9.0-11.8% across the three benchmarks) while reducing retrieval calls by 47% compared to fixed-interval approaches like IRCoT (all improvements significant at p<0.01, paired bootstrap). On the challenging MuSiQue benchmark requiring 2-4 hop reasoning, our method achieves 71.2% F1 with an average of only 1.8 retrieval calls per question. Analysis shows that ReaLM-Retrieve also improves retrieval quality itself, achieving 81.3% Recall@5 with consistently higher precision and MRR than fixed-interval baselines on supporting evidence, establishing new state-of-the-art efficiency-accuracy trade-offs for reasoning-intensive retrieval tasks.
Dongxin Guo, Jikun Wu, Siu Ming Yiu
The University of Hong Kong Hong Kong, China · Brain Investing Limited Hong Kong, China · Stellaris AI Limited Hong Kong, China
As the world changes, new information is published while older records remain online. This complicates reasoning for large language models (LLMs), which must retrieve evidence and resolve conflicts to determine which information remains valid. Existing evaluations treat information seeking as a black box, revealing little about where and why LLMs fail. To address this gap, we introduce TREK, a suite of three components: TREKTRACE is a plug-and-play framework that attributes failures to specific stages of retrieval and reasoning. We apply it to TREKBENCH-WILD, a meta-benchmark of real-world questions that require complex reasoning over rapidly evolving knowledge. Surprisingly, the main bottleneck is post-retrieval reasoning rather than retrieval: frontier models are often misled by conflicting evidence even after retrieving all the information needed to answer. To study this bottleneck, we build TREKBENCH-CTRL, a controlled benchmark with distractors spanning temporality, relevance, and trustworthiness, and show that benchmarks dominated by outdated distractors substantially underestimate task difficulty. Together, TREK provides a testbed for diagnosing and improving reasoning over rapidly evolving knowledge.
Tianyi Hu, Bhavyajeet Singh, Niket Tandon +1
Aarhus University · Microsoft Research · Technische Universität Darmstadt