IterSynth: Rethinking Deep Search Agents via Role-Decoupled Iterative Synthesis

Hugging Face Daily Papers Papers
deep-search-agents iterative-synthesis role-decoupling reinforcement-learning summarization LLM-agents benchmarks

Summary

IterSynth introduces a role-decoupled iterative synthesis paradigm for deep search agents, using reinforcement learning to improve performance on long-horizon search tasks and surpassing prior methods on benchmarks.

Deep search requires LLM agents to decompose complex queries, search for evidence, and synthesize grounded answers, yet existing ReAct-style agents suffer from two limitations: role coupling, where one policy must handle planning, evidence use, and synthesis; and context accumulation, where growing search histories introduce noise and obscure useful information. To address these issues, we propose IterSynth, a role-decoupled and summary-based paradigm that alternates between a Planner for identifying information needs and a Synthesizer for integrating evidence into an evolving summary state. This design separates planning from synthesis while using the summary as the persistent state of search, reducing both capability coupling and context noise. To train IterSynth effectively, we further introduce Role-Decoupled Policy Optimization (RDPO) for reinforcement learning, which combines terminal outcome rewards with turn-level rubric evaluations and computes role-specific advantages for more precise credit assignment. Experiments on five long-horizon deep-search benchmarks such as BrowseComp and Xbench-DS show that IterSynth-8B achieves an average score of 50.7, surpassing the strongest prior leq8B agent by +4.2\%. Moreover, IterSynth serves as a model-agnostic prompting paradigm, delivering substantial zero-shot gains over ReAct and similar prompting paradigms on frontier proprietary models.
Original Article
View Cached Full Text

Cached at: 09/25/26, 03:43 AM

Paper page - IterSynth: Rethinking Deep Search Agents via Role-Decoupled Iterative Synthesis

Source: https://huggingface.co/papers/2609.29444 Authors:

,

,

,

,

,

,

,

,

,

,

Abstract

DeepsearchrequiresLLMagentstodecomposecomplexqueries,searchforevidence,andsynthesizegroundedanswers,yetexistingReAct-styleagentssufferfromtwolimitations:rolecoupling,whereonepolicymusthandleplanning,evidenceuse,andsynthesis;andcontextaccumulation,wheregrowingsearchhistoriesintroducenoiseandobscureusefulinformation.Toaddresstheseissues,weproposeIterSynth,arole-decoupledandsummary-basedparadigmthatalternatesbetweenaPlannerforidentifyinginformationneedsandaSynthesizerforintegratingevidenceintoanevolvingsummarystate.Thisdesignseparatesplanningfromsynthesiswhileusingthesummaryasthepersistentstateofsearch,reducingbothcapabilitycouplingandcontextnoise.TotrainIterSyntheffectively,wefurtherintroduceRole-DecoupledPolicyOptimization(RDPO)forreinforcementlearning,whichcombinesterminaloutcomerewardswithturn-levelrubricevaluationsandcomputesrole-specificadvantagesformoreprecisecreditassignment.Experimentsonfivelong-horizondeep-searchbenchmarkssuchasBrowseCompandXbench-DSshowthatIterSynth-8Bachievesanaveragescoreof50.7,surpassingthestrongestpriorleq8Bagentby+4.2\%.Moreover,IterSynthservesasamodel-agnosticpromptingparadigm,deliveringsubstantialzero-shotgainsoverReActandsimilarpromptingparadigmsonfrontierproprietarymodels.

View arXiv pageView PDFGitHub2Add to collection

Get this paper in your agent:

hf papers read 2609\.29444

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2609.29444 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2609.29444 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2609.29444 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

RODS: Reward-Driven Online Data Synthesis for Multi-Turn Tool-Use Agents

arXiv cs.AI

This paper introduces RODS, a reward-driven online data synthesis method that addresses the depletion of informative samples in static datasets for multi-turn tool-use agent training. It achieves comparable performance to larger offline pipelines with significantly fewer trajectories.