WebSwarm: Recursive Multi-Agent Orchestration for Deep-and-Wide Web Search

# WebSwarm: Recursive Multi-Agent Orchestration for Deep-and-Wide Web Search
Source: [https://arxiv.org/html/2607.08662](https://arxiv.org/html/2607.08662)
Xiaoshuai Song1, Liancheng Zhang1∗\*, Kangzhi Zhao2\\corresponding, Yutao Zhu1, Zhongyuan Wang2∗\*, Guanting Dong1, Jinghan Yang2∗\*, Han Li2, Kun Gai2, Ji\-Rong Wen1, Zhicheng Dou1\\corresponding

###### Abstract

Large language model \(LLM\)\-based web search agents are transforming information seeking from simple factoid question answering into complex, deep\-and\-wide search and research\-oriented tasks\. A single ReAct\-style agent is constrained by one long trajectory and limited context, making it difficult to handle depth and coverage simultaneously\. Existing multi\-agent systems improve search coverage through parallel execution and aggregation, but still exhibit clear limitations in recursive depth, collaboration adaptability, and evidence\-grounded expansion\. We propose WebSwarm, a progressive recursive delegation framework that jointly constructs task decomposition, recursive expansion, and agent collaboration during inference\. WebSwarm dynamically instantiates agentic search nodes, each coupling a local objective with a search mode that specifies how the node should organize search and collaboration\. Each node can either solve its objective itself or further delegate child nodes; after solving, it returns evidence and results upward, enabling parent nodes to further expand, revise, or aggregate the search process\. To guide this process, WebSwarm first probes how task\-relevant information is organized on the web to ground subsequent node expansion, and reuses process\-level experience across homogeneous sibling nodes\. Experiments on BrowseComp\-Plus, WideSearch, DeepWideSearch, and GISA show that WebSwarm consistently outperforms single\-agent and multi\-agent baselines on deep, wide, and interleaved deep\-and\-wide tasks\. Further analyses of ablation, task difficulty, web tool efficiency, and model generalization explain WebSwarm’s effectiveness and provide insights for multi\-agent search systems111Github:https://github\.com/songxiaoshuai/WebSwarm\.

## Introduction

![Refer to caption](https://arxiv.org/html/2607.08662v1/x1.png)Figure 1:Illustration of representative multi\-agent orchestration paradigms and WebSwarm\.The development of large language models \(LLMs\) is driving web information seeking toward agentic search, enabling search agents to autonomously perform multi\-turn search and web browsing to gather information for user queries\(Zhuet al\.[2026a](https://arxiv.org/html/2607.08662#bib.bib18); Xiet al\.[2025](https://arxiv.org/html/2607.08662#bib.bib28)\)\. As this paradigm evolves, search agents are moving beyond simple factoid QA toward more complex information\-seeking tasks, such as research\-level search and report\-oriented investigation\. This requires agents to support both deep and wide search: deep search resolves multi\-hop dependencies and constraints, while wide search maintains sufficient coverage across candidate entities, web pages, and information sources\. Recently, a series of benchmarks have evaluated the capability boundaries of search agents from the perspectives of depth, width, and their nested interaction\(Weiet al\.[2025](https://arxiv.org/html/2607.08662#bib.bib3); Chenet al\.[2025b](https://arxiv.org/html/2607.08662#bib.bib4); Wonget al\.[2026](https://arxiv.org/html/2607.08662#bib.bib2); Lanet al\.[2025](https://arxiv.org/html/2607.08662#bib.bib5); Zhuet al\.[2026b](https://arxiv.org/html/2607.08662#bib.bib1)\)\.

Both benchmark results and practical experience show that a single ReAct\(Yaoet al\.[2023](https://arxiv.org/html/2607.08662#bib.bib16)\)agent struggles as task depth and width increase\. To address this issue, recent studies have introduced multi\-agent search systems\(Jinet al\.[2025b](https://arxiv.org/html/2607.08662#bib.bib6); Team[2026](https://arxiv.org/html/2607.08662#bib.bib7); Lanet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib9); Chenet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib10); Alzubiet al\.[2026a](https://arxiv.org/html/2607.08662#bib.bib13); Leeet al\.[2026a](https://arxiv.org/html/2607.08662#bib.bib12); Ninget al\.[2026](https://arxiv.org/html/2607.08662#bib.bib14)\)\. These systems assign work to multiple agents and use mechanisms such as parallel search, cross\-checking, and result aggregation to improve coverage and reliability\.

For complex web search tasks with intertwined depth and width, the solving process is often difficult to fully determine from the initial query alone; instead, intermediate search evidence progressively reveals new entities, constraints, and goals\. Therefore, rather than committing upfront to a fixed subtask decomposition and agent\-collaboration structure,an ideal multi\-agent system should progressively and recursively construct both as evidence accumulates\. However, as illustrated in Figure[1](https://arxiv.org/html/2607.08662#Sx1.F1), existing multi\-agent search systems still exhibit clear limitations in recursive depth, collaboration adaptability, and evidence\-grounded expansion:

\(1\) Shallow recursive depth\.Deeper subtasks may only become clear after earlier subtasks are expanded or solved, and some subtasks may introduce new dependencies and requirements\. Thus, the task tree should support progressive and recursive expansion\. Existing systems usually decompose only at the root level\. When a task requires multi\-level expansion, such as “year→\\rightarrowbrand→\\rightarrowmodel→\\rightarrowattribute”, deeper structures are forced into a long ReAct trajectory of a sub\-agent, making the process close to single\-agent search\.

\(2\) Limited collaboration adaptability\.As the search process recursively unfolds, local search nodes may expose different goals and bottlenecks: fact lookup requires fast and reliable execution, wide search requires coverage over many items, deep search requires iterative clue discovery and verification, and open\-set enumeration requires balancing recall and precision under unknown set boundaries\. However, existing methods typically rely on a single global collaboration paradigm, such as multi\-sample aggregation, serial handoff, or parallel divide\-and\-conquer\. Although each paradigm is effective in certain search scenarios, a fixed collaboration pattern struggles to cover all search needs or flexibly switch collaboration forms across different subtasks\.

\(3\) Weakly evidence\-grounded expansion\.Required information may be concentrated in a few aggregate pages, or distributed across timelines, entities, event sets, or attribute dimensions\. A proper decomposition should align with the actual organization of web information\. Therefore, the system must decide how to expand with respect to how relevant information is organized on the web\. However, existing multi\-agent systems often split tasks only by surface semantics of the query\. When the decomposition dimension is not aligned with the web information structure, the system may over\-decompose concentrated information or split dispersed information along the wrong dimension, leading to redundant retrieval, insufficient recall, and difficult aggregation\.

To address these limitations, we propose WebSwarm, a progressive recursive multi\-agent framework that organizes complex web information seeking by dynamically creating and delegating search nodes with diverse local objectives and collaboration patterns\. In WebSwarm, the root agent receives the original task, creates and delegates search nodes\. Each search node is itself an agent, receiving a local objective and a search mode\. The search mode determines how the node solves its local objective: either by conducting iterative search on its own, or by recursively generating and delegating child nodes while organizing corresponding multi\-node collaboration structures, such as parallel divide\-and\-conquer, sequential search\-and\-verification, or multi\-path sampling and aggregation\. Once a node completes its local objective, it returns the result upward, and the upper\-level agent decides whether to further expand, revise, or terminate the search process based on the returned information\. In this way, WebSwarm unifies recursive delegation, multi\-level feedback, and diverse local collaboration structures into a progressive search process\. To avoid blind recursive delegation, WebSwarm further introduces two complementary signals: external web information structure and internal experience\. On the one hand, the system performs lightweight probing of web information structure to determine whether relevant information is concentrated in a few aggregated pages or dispersed along a certain organizational dimension, thereby guiding how subsequent search nodes should be expanded\. On the other hand, for homogeneous search nodes under the same parent node, WebSwarm distills trajectory experience from a small number of preceding nodes to guide the local search of subsequent nodes\.

We evaluate WebSwarm on four challenging web information seeking benchmarks: BrowseComp\-Plus, WideSearch, DeepWideSearch, and GISA, covering deep, wide, and hybrid deep\-wide search tasks\. Experimental results show that WebSwarm consistently outperforms ReAct agent and multi\-agent baselines\. Further analyses examine module ablations, the relationship between task difficulty and method performance, web tool\-use efficiency, and generalization across different models, providing insights into WebSwarm’s effectiveness and the design of multi\-agent search systems\.

Overall, we propose WebSwarm for deep and wide web search tasks\. Our contributions are threefold:

- •We propose a progressive recursive delegation framework that instantiates agentic search nodes coupling local objectives with search modes, jointly constructing decomposition, expansion, and collaboration\.
- •We introduce web\-structure\-guided recursive delegation, grounding expansion in web information organization and further transferring subtask experience across homogeneous search nodes\.
- •Experiments on four benchmarks verify WebSwarm’s effectiveness on complex web information seeking tasks involving deep, wide, and depth\-width interleaved search\.

![Refer to caption](https://arxiv.org/html/2607.08662v1/x2.png)Figure 2:Overview of WebSwarm with a running example from the DeepWideSearch benchmark\.
## Related Work

### Agent for Web Information Seeking

Agentic search combines LLMs with web search and browsing tools to solve information\-seeking tasks\(Nakanoet al\.[2021](https://arxiv.org/html/2607.08662#bib.bib17); Zhuet al\.[2026a](https://arxiv.org/html/2607.08662#bib.bib18); Liet al\.[2025b](https://arxiv.org/html/2607.08662#bib.bib19); Xiet al\.[2025](https://arxiv.org/html/2607.08662#bib.bib28)\)\. One line of work focuses on model training to internalize information\-seeking strategies\(Jinet al\.[2025a](https://arxiv.org/html/2607.08662#bib.bib29); Liet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib21); Wuet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib22); Liet al\.[2025a](https://arxiv.org/html/2607.08662#bib.bib23)\)\. Recent benchmarks including BrowseComp\(Weiet al\.[2025](https://arxiv.org/html/2607.08662#bib.bib3)\), WideSearch\(Wonget al\.[2026](https://arxiv.org/html/2607.08662#bib.bib2)\), DeepWideSearch\(Lanet al\.[2025](https://arxiv.org/html/2607.08662#bib.bib5)\), and GISA\(Zhuet al\.[2026b](https://arxiv.org/html/2607.08662#bib.bib1)\)reveal the difficulties that agents face in deep and wide web search tasks\. To address these challenges, inference\-time architectures organize complex search through explicit state structures and multi\-agent collaboration\(Qinet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib8); Xuet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib11); Ninget al\.[2026](https://arxiv.org/html/2607.08662#bib.bib14)\)\. Table\-as\-Search\(Lanet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib9)\)and Web2BigTable\(Huanget al\.[2026](https://arxiv.org/html/2607.08662#bib.bib30)\)formulate search as table completion\. MindSearch\(Chenet al\.[2025a](https://arxiv.org/html/2607.08662#bib.bib20)\)and HiRA\(Jinet al\.[2025b](https://arxiv.org/html/2607.08662#bib.bib6)\)adopt graph\-based or hierarchical planning\. InfoSeeker\(Leeet al\.[2026a](https://arxiv.org/html/2607.08662#bib.bib12)\)and A\-MapReduce\(Chenet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib10)\)emphasize hierarchical parallelism or MapReduce\-style horizontal decomposition\. In contrast, WebSwarm formulates web search as evidence\-driven recursive delegation: it instantiates search nodes as intermediate evidence reveals new local objectives, grounds their expansion in web information structure, and couples each objective with a search mode to handle heterogeneous local search needs\.

### Multi\-Agent Systems for Long\-Horizon Tasks

Early multi\-agent systems, including CAMEL\(Liet al\.[2023](https://arxiv.org/html/2607.08662#bib.bib35)\), AutoGen\(Wuet al\.[2023](https://arxiv.org/html/2607.08662#bib.bib32)\), MetaGPT\(Honget al\.[2024](https://arxiv.org/html/2607.08662#bib.bib33)\), and ChatDev\(Qianet al\.[2024](https://arxiv.org/html/2607.08662#bib.bib34)\), explore agent collaboration from perspectives such as role\-playing, software development, debate, and voting\. Recent work further investigates agent orchestration for long\-horizon tasks\. Claude Agent Team builds agent teams that can communicate and collaborate with each other\(Liuet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib27)\); and Kimi\-Swarm decomposes complex tasks into multiple subproblems and executes multiple worker agents in parallel\(Team[2026](https://arxiv.org/html/2607.08662#bib.bib7)\)\. In addition, Magentic\-One employs a central orchestrator to coordinate specialized agents\(Fourneyet al\.[2024](https://arxiv.org/html/2607.08662#bib.bib24)\); ROMA\(Alzubiet al\.[2026a](https://arxiv.org/html/2607.08662#bib.bib13)\)represents long\-horizon tasks as recursive subtask trees; AgentFugue\(Huet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib31)\)reuses intermediate findings from parallel agents through a shared reasoning hub; and AggAgent\(Leeet al\.[2026b](https://arxiv.org/html/2607.08662#bib.bib36)\)enables test\-time scaling by aggregating multiple long\-horizon agent trajectories\. In contrast, WebSwarm views multi\-agent orchestration as recursive delegation rather than a fixed global collaboration topology\. It recursively instantiates agentic nodes whose modes define local collaboration protocols; each node may solve locally, delegate child nodes, and return evidence upward, allowing decomposition and collaboration to co\-evolve during inference\.

## Methodology

### Preliminaries and Overview

##### ReAct Searcher\.

For a web information\-seeking taskq0q\_\{0\}, the ReAct agent searches and reads web pages through web tools at each step until it returns the final answeraa\. The tool actions includesearch\(query\)andfetch\_url\(url\): the former returns the most relevant webpage URLs and snippets for the query, while the latter returns the text content of a specified webpage\.

##### WebSwarm\.

As shown in Figure[2](https://arxiv.org/html/2607.08662#Sx1.F2), at the core of WebSwarm is a recursive delegation process, where the system progressively solves the original task by dynamically creating search nodes and enabling delegation and feedback among them\. Each search node is itself an agent, receiving a local objective and a search mode, where the search mode determines whether the node directly solves the current objective or further delegates child nodes and organizes local collaboration structures\. The results returned by child nodes then serve as the basis for upper\-level nodes to further expand, revise, or aggregate the search process \(§3\.2\)\. To make recursive delegation more reliable and efficient, WebSwarm further leverages two types of guiding signals: probed external web information structure guides how subsequent search nodes are expanded, while internal process experience guides the subsequent search of homogeneous nodes \(§3\.3\)\.

### Recursive Search Delegation

Given the original taskq0q\_\{0\}, WebSwarm dynamically constructs a recursive delegation tree𝒯=\(𝒱,ℰ\)\\mathcal\{T\}=\(\\mathcal\{V\},\\mathcal\{E\}\)during execution\. The root nodev0v\_\{0\}receivesq0q\_\{0\}and serves as the starting point of recursive delegation\. Each non\-root nodev∈𝒱v\\in\\mathcal\{V\}corresponds to an agent and receives a local objectiveqvq\_\{v\}and a search modemvm\_\{v\}:qvq\_\{v\}specifies what the node should solve, whilemvm\_\{v\}specifies how it organizes search and collaboration\. An edge\(u,v\)∈ℰ\(u,v\)\\in\\mathcal\{E\}denotes one delegation from parent nodeuuto child nodevv\. During solving, nodevvmaintains the set of resultsRvR\_\{v\}returned by its children, and returns resultrvr\_\{v\}to its parent after completing its local objective\.

#### Objective Decomposition and Delegation\.

The basic action in recursive delegation is that a node creates one or more child nodes\. For a nodevv, the agent determines whether to expand further based on the local objectiveqvq\_\{v\}, search modemvm\_\{v\}, and collected child resultsRvR\_\{v\}\. If the current evidence is insufficient, nodevvgenerates a set of child delegations:

Cv=\{\(qi,mi\)\}i=1nv\.C\_\{v\}=\\\{\(q\_\{i\},m\_\{i\}\)\\\}\_\{i=1\}^\{n\_\{v\}\}\.\(1\)Each child delegation specifies both a local objectiveqiq\_\{i\}and a corresponding search modemim\_\{i\}, instantiating a new search nodeviv\_\{i\}and forming an edge\(v,vi\)\(v,v\_\{i\}\)in the recursive delegation tree\. The key point is that WebSwarm generates not merely a set of subtasks, but a set of objective–mode pairs\. Since different local objectives often require different solution strategies, coupling the objective with its search mode during delegation ensures each downstream node has a solving protocol and collaboration structure suited to its goal\.

#### Search Node Solving\.

The search mode specifies the solving protocol for the subtaskqvq\_\{v\}\. By assigning different instructions, roles, and tools to a node, the search mode determines whether the node directly calls web tools, or further creates child nodes and organizes collaboration among them\. WebSwarm defines four search modes:

ℳ=\{atom,deep,wide,entity\_collect\}\.\\mathcal\{M\}=\\\{\\textit\{atom\},\\textit\{deep\},\\textit\{wide\},\\textit\{entity\\\_collect\}\\\}\.\(2\)They correspond to four subtask bottlenecks in web search:

- •atom: atomic fact lookup\.atomis the basic search mode, suited to focused, narrow local queries\. Its goal is to quickly and reliably find evidence\. Such a node corresponds to a ReAct agent that directly calls web tools to locate relevant evidence and return the result\.
- •deep: iterative search and verification\.deepapplies to unknown\-target identification or multi\-constraint reasoning\. The challenge is not covering many objects, but iteratively proposing candidates from indirect clues, verifying them, and adjusting the search direction\. Thus,deepuses a serial search–verification structure: iteratively dispatches searchers and verifiers\. Searchers propose candidates or evidence from different clue paths, while verifiers independently check whether candidates satisfy the constraints\. Each searcher or verifier is anatomnode\.
- •wide: parallel divide\-and\-conquer\.wideis suited to collecting similar information over a set of objects or dimensions, such as filling attributes for multiple entities\. The bottleneck is coverage and structured aggregation, sowideconcurrently dispatches a set of independent child nodes and merges their results\. Its children are not limited toatom: in tasks combining depth and breadth, child nodes can also bedeep,entity\_collect, orwide, enabling nested information collection\.
- •entity\_collect: multi\-path recall and verification\.entity\_collectapplies when the set boundary is unknown and the task is to enumerate a complete set of entities\. Unlike attribute filling over known objects, its core challenge is balancing recall and precision: a single search path may miss members, while broadening the search directly may introduce noise\. Therefore,entity\_collectconcurrently dispatches multipleatomsearch nodes to recall candidate entities from different sources or perspectives, and obtains the target set by merging, deduplicating, and verifying low\-confidence candidates\.

Through these four search modes, WebSwarm enables each search node to choose an appropriate solving protocol for its local objective and combine multiple collaboration structures within the same recursive process\.

#### Evidence Update and Continuation\.

When a child nodeviv\_\{i\}completes its local objective, it returns the resultrir\_\{i\}to its parent nodevv, which updates its evidence state:

Rv←Rv∪\{ri\}\.R\_\{v\}\\leftarrow R\_\{v\}\\cup\\\{r\_\{i\}\\\}\.\(3\)Based on these returned resultsRvR\_\{v\}, the parent decides whether to continue or stop\. If the local objective remains unresolved, it may generate new sub\-delegations, revise the search direction, or request finer\-grained evidence\. Otherwise, it aggregates the collected results and returns upward:

rv=Aggregate​\(qv,Rv\)\.r\_\{v\}=\\textsc\{Aggregate\}\(q\_\{v\},R\_\{v\}\)\.\(4\)Thus, returned results serve not only as answer components but also as control signals for further delegation\. WebSwarm therefore alternates between top\-down delegation and bottom\-up evidence feedback, allowing the search structure to unfold progressively during execution\.

### Guiding Delegation with Web Structure and Experience

During recursive delegation, agent nodes may generate misaligned expansions, redundant delegations, or repeated trial\-and\-error during search\. To improve reliability and efficiency, WebSwarm introduces two complementary guidance signals: web\-structure cueshvh\_\{v\}from lightweight probing, which constrain the scope, granularity, and dimensions of subsequent node expansion, and internal search experiencekvk\_\{v\}, which guides the solving process of homogeneous nodes\.

#### Web\-Structure\-Guided Expansion\.

Recursive expansion requires deciding not only whether to generate more search nodes, but also along which dimension to expand\. Prior methods mainly rely on LLM agents to decompose tasks based on query surface semantics, which can be misaligned with the actual organization of web evidence in two ways: \(1\) when the required information is concentrated in a few aggregate pages, premature expansion by entity or attribute causes redundant retrieval; \(2\) when information is dispersed across sources organized by external structures such as time, events, organizations, or entities, choosing the wrong expansion dimension reduces efficiency and increases aggregation noise\.

To mitigate this misalignment, WebSwarm introduces a Web\-Probing Agent for nodes that require wide expansion\. It acts as a lightweight pre\-expansion structure scout, probes how task\-relevant evidence is organized on the web and uses this structure signal to guide subsequent delegation\. Given the current local objectiveqvq\_\{v\}, the Web\-Probing Agent performs lightweight search and webpage reading, and returns a web\-structure hint:

hv=WebProbing​\(qv\)\.h\_\{v\}=\\textsc\{WebProbing\}\(q\_\{v\}\)\.\(5\)Here,hvh\_\{v\}summarizes the preliminarily observed evidence distribution, representative pages and their supporting signals, and possible expansion axes\. The node then generates child delegations conditioned onhvh\_\{v\}:

Cv=Delegate​\(qv,mv,Rv,hv\)\.C\_\{v\}=\\textsc\{Delegate\}\(q\_\{v\},m\_\{v\},R\_\{v\},h\_\{v\}\)\.\(6\)When information is concentrated in a few aggregate pages, WebSwarm creates a small number of extraction\-oriented search nodes around these pages to avoid redundant parallelism\. When evidence is dispersed across multiple sources, it instead expands along the web\-organization axes suggested byhvh\_\{v\}, improving coverage and aggregability\.

#### Experience\-Guided Node Solving\.

During recursive delegation, thewidenode often creates a batch of homogeneous search nodes, such as collecting attributes for a set of known entities\. Although these nodes target different entities, they often share similar query patterns, reliable sources, page structures, and failure paths\. If each node searches in isolation, it may repeat unreliable strategies, miss useful source patterns, or produce inconsistent results across sibling nodes\. To address this, WebSwarm introduces within\-instance subtask experience transfer: it first executes a small number of scout nodes, extracts process\-level subtask experience from their search trajectories, and uses it to guide subsequent homogeneous nodes toward more reliable solving\. Specifically, given a set of homogeneous child delegationsCvC\_\{v\}under nodevv, WebSwarm first executes a small scout subsetCvs⊂CvC\_\{v\}^\{s\}\\subset C\_\{v\}:

\(ri,τi\)=SolveNode​\(vi\),\(qi,mi\)∈Cvs\.\(r\_\{i\},\\tau\_\{i\}\)=\\textsc\{SolveNode\}\(v\_\{i\}\),\\quad\(q\_\{i\},m\_\{i\}\)\\in C\_\{v\}^\{s\}\.\(7\)whereviv\_\{i\}is the search node instantiated by the child delegation\(qi,mi\)\(q\_\{i\},m\_\{i\}\), andτi\\tau\_\{i\}denotes its solving trajectory\. WebSwarm then extracts subtask experiencekvk\_\{v\}from these trajectories, including useful query patterns, reliable sources, page types, and invalid paths:

kv=ExtractExperience​\(\{τi:\(qi,mi\)∈Cvs\}\)\.k\_\{v\}=\\textsc\{ExtractExperience\}\(\\\{\\tau\_\{i\}:\(q\_\{i\},m\_\{i\}\)\\in C\_\{v\}^\{s\}\\\}\)\.\(8\)For the remaining sibling nodes, WebSwarm injectskvk\_\{v\}into their solving context:

rj=SolveNode​\(vj;kv\),\(qj,mj\)∈Cv∖Cvs\.r\_\{j\}=\\textsc\{SolveNode\}\(v\_\{j\};k\_\{v\}\),\\quad\(q\_\{j\},m\_\{j\}\)\\in C\_\{v\}\\setminus C\_\{v\}^\{s\}\.\(9\)Different from studies focusing on cross\-task experience or skills\(Chenet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib10); Alzubiet al\.[2026b](https://arxiv.org/html/2607.08662#bib.bib15)\), this mechanism only reuses experience within the same instance and among homogeneous sibling nodes under the same parent, thereby preserving evaluation\-sample independence\.

MethodBackboneBC\-PlusWideSearch\-ENDeepWideSearch\-ENGISAACCSRRow F1Item F1SRRow F1Item F1ItemSetListTableOverallSingle\-AgentReActKimi\-K266\.507\.0037\.4267\.653\.9527\.5853\.5718\.1857\.1848\.7658\.3454\.58ReActQwen3\.5\-35B56\.505\.0037\.3864\.823\.9528\.7952\.4627\.2755\.7651\.7256\.0353\.74ReActQwen3\-235B21\.503\.0020\.8949\.150\.0011\.7536\.1240\.9152\.3736\.4843\.9345\.57ReActGLM\-4\.550\.504\.0033\.2364\.613\.9520\.0846\.6327\.2751\.2750\.5859\.7855\.54Multi\-AgentSwarm\-AgentGLM\-4\.564\.506\.0036\.6668\.791\.3227\.0351\.4131\.8254\.4352\.3360\.6657\.05Flash\-SearcherGLM\-4\.554\.005\.0036\.5368\.685\.2625\.1750\.0018\.1854\.9156\.8556\.7254\.22Table\-as\-SearchGLM\-4\.562\.506\.0037\.9769\.443\.9523\.1254\.9622\.7360\.4457\.8660\.8258\.14ROMAGLM\-4\.542\.505\.0033\.4267\.192\.6324\.0250\.5627\.2756\.9156\.4660\.5557\.57InfoSeekerGLM\-4\.559\.504\.0039\.9871\.912\.6325\.8155\.1027\.2754\.5857\.3140\.8258\.99\\rowcolorrowblue\[2\]\[2\]WebSwarmGLM\-4\.568\.00\(\+17\.50\)7\.00\(\+3\.00\)44\.14\(\+10\.91\)74\.37\(\+9\.76\)6\.58\(\+2\.63\)29\.64\(\+9\.56\)58\.40\(\+11\.77\)40\.91\(\+13\.64\)61\.03\(\+9\.76\)66\.04\(\+15\.46\)63\.69\(\+3\.91\)62\.30\(\+6\.76\)Table 1:Performance comparison of methods across four benchmarks\. The top two results are highlighted in bold and underlined\. Numbers in parentheses denote the gain of WebSwarm compared to the ReAct agent under GLM\-4\.5\. BC\-Plus: BrowseComp\-Plus; SR: Success Rate; Qwen3\-235B: Qwen3\-235B\-A22B\-2507; Kimi\-K2: Kimi\-K2\-Thinking\.

## Experiment

### Experimental Setup

#### Benchmarks\.

We evaluate WebSwarm on four challenging web information\-seeking benchmarks: BrowseComp\-Plus\(Chenet al\.[2025b](https://arxiv.org/html/2607.08662#bib.bib4)\)for deep factual search, WideSearch\(Wonget al\.[2026](https://arxiv.org/html/2607.08662#bib.bib2)\)for structured wide information collection, DeepWideSearch\(Lanet al\.[2025](https://arxiv.org/html/2607.08662#bib.bib5)\)for nested deep and wide search, and GISA\(Zhuet al\.[2026b](https://arxiv.org/html/2607.08662#bib.bib1)\)for general information seeking\. For resource efficiency, we randomly sample 200 instances from BrowseComp\-Plus and use the English subsets of WideSearch and DeepWideSearch\.

Following the official evaluation settings, we report answer accuracy \(ACC\) for BrowseComp\-Plus\. For WideSearch and DeepWideSearch, whose outputs are tables, we report item\-F1, row\-F1, and success rate \(SR\) to measure cell\-level, row\-level, and full\-table correctness, respectively\. GISA covers diverse task types and metrics, including exact match \(EM\) for items, set F1, list F1, and table item\-F1\.

#### Baselines\.

We compare WebSwarm with the following representative search agent and multi\-agent methods: ReAct agent\(Yaoet al\.[2023](https://arxiv.org/html/2607.08662#bib.bib16)\); Swarm\-Agent, which follows the Kimi\-Swarm\(Team[2026](https://arxiv.org/html/2607.08662#bib.bib7)\)and dynamically creates and assigns tasks to subagents; Flash\-Searcher\(Qinet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib8)\), which organizes subtasks as a dynamic dependency graph; Table\-as\-Search\(Lanet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib9)\), which represents search progress as table completion; InfoSeeker\(Leeet al\.[2026a](https://arxiv.org/html/2607.08662#bib.bib12)\), which adopts a hierarchical Host–Manager–Worker architecture; and ROMA\(Alzubiet al\.[2026a](https://arxiv.org/html/2607.08662#bib.bib13)\), which recursively decomposes tasks and aggregates leaf\-level results\.

#### Implementation\.

We mainly use GLM\-4\.5 as the backbone model for WebSwarm and multi\-agent baselines, and compare different models in Section 4\.5\. For web interaction, all methods use the same Web\-Search and Page\-Browse tools\. For WebSwarm, we set the number of parallel search paths inentity\_collectto 3, and use 2 scout child nodes for experience extraction\. We use each model’s maximum context length and allow up to 200 action steps to avoid prematurely truncating agents\. We provide detailed descriptions of the benchmarks, baselines, implementation in Appendix\.

### Main Results

As shown in Table[1](https://arxiv.org/html/2607.08662#Sx3.T1), WebSwarm achieves the best or competitive results across four benchmarks, consistently outperforming both single\-agent ReAct and multi\-agent baselines\. Next, we analyze the results from three aspects:

\(1\) Deep Search\.On BrowseComp\-Plus, WebSwarm improves over ReAct by 17\.50 accuracy points and outperforms the strongest multi\-agent baseline by 3\.50 points\. This gain is primarily attributed to thedeepsearch mode, where the searcher agent iteratively shifts exploration perspectives and refines candidate hypotheses in a sequential manner, while an independent verifier adversarially evaluates and filters candidate answers, leading to deeper web information exploration\.

\(2\) Wide Search\.On WideSearch\-EN, WebSwarm achieves gains of 10\.91 and 9\.76 points in Row F1 and Item F1 over ReAct, and also shows improvements over multi\-agent baselines\. These results show that WebSwarm is effective not only at expanding search coverage, but also at aggregating structured evidence with higher row\- and item\-level correctness\. These gains stem from recursive delegation amongentity\_collect,wide, andatomsearch nodes, further supported by Web\-structure\-guided expansion and process\-level experience reuse across homogeneous sibling nodes\.

\(3\) Interleaved Deep and Wide Search\.WebSwarm improves over ReAct by 9\.56 Row F1 and 11\.77 Item F1 on DeepWideSearch\-EN\. On GISA, WebSwarm performs strongly across all task subsets\. These results indicate that when tasks require alternating between deep and wide exploration, or combining both within a single workflow, fixed collaboration patterns may be less effective at coordinating different search phases\. WebSwarm addresses this through a recursive delegation structure and mode\-guided node solving, allowing flexible transitions between deep and wide search as evidence accumulates\.

MethodBC\-PlusWideSearchDeepWideSearchACCItem F1Item F1WebSwarm68\.0074\.3758\.40w/o recursive63\.5068\.3855\.79All\-to\-wide63\.0072\.0155\.87All\-to\-deep67\.5069\.9454\.51Table 2:Ablation of recursive delegation and search mode\.MethodWideSearchDeepWideSearchItem F1Web ToolItem F1Web ToolWebSwarm74\.37137\.0358\.40203\.73w/o Web\-Probing74\.90239\.9058\.93331\.39w/o Experience71\.20132\.5755\.48220\.07Table 3:Ablation of Web\-probing and experience reuse\.![Refer to caption](https://arxiv.org/html/2607.08662v1/x3.png)Figure 3:Performance and web tool usage of GLM\-4\.5 across difficulty levels on BrowseComp\-Plus and WideSearch\-EN\.
### Ablation Analysis

Structure Ablation\.To analyze the role of recursive delegation and search mode, we construct three variants:w/o Recursive Delegation, which allows the root node to create child nodes but prevents child nodes from further delegation; and All\-to\-wide/ All\-to\-deep, which force all non\-atomnodes to use a single search mode\. As shown in Table[2](https://arxiv.org/html/2607.08662#Sx4.T2), removing recursive delegation consistently degrades performance, confirming the value of progressively refined search structures\. The single\-mode variants show task\-dependent behavior: All\-to\-deepremains close to WebSwarm on BC\-Plus, but drops clearly on WideSearch and DeepWideSearch; conversely, All\-to\-wideis less harmful on WideSearch but substantially weakens BC\-Plus\. It highlights the importance of matching local objectives with suitable search modes\.

Guidance Signal Ablation\.Since web\-structure probing and experience reuse mainly affect expansion and local solving underwidenodes, we conduct ablations on WideSearch and DeepWideSearch in Table[3](https://arxiv.org/html/2607.08662#Sx4.T3)\. Removing Web\-Probing causes only minor changes in Item F1, but substantially increases the average number of Web tool calls from 137\.03 to 239\.90 and from 203\.73 to 331\.39, respectively\. This suggests that Web\-Probing primarily improves search efficiency by reducing redundant or misaligned node expansion\. In contrast, removing experience reuse consistently lowers Item F1 on both datasets, indicating that process\-level experience improves the reliability of homogeneous sibling nodes\. Overall, Web\-Probing and experience reuse play complementary roles: the former reduces exploration cost, while the latter improves node\-solving quality\.

### Task Difficulty and Tool Usage Analysis

To investigate WebSwarm’s performance gains across task difficulties and its web\-tool efficiency, we use ReAct Agent as the anchor for difficulty stratification\. For BrowseComp\-Plus, difficulty is defined by ReAct Agent’s pass rate over three samples: Easy means all samples pass, Hard means all fail, and Mid denotes the remaining cases\. For WideSearch\-EN, samples are sorted by ReAct Agent’s Item F1 score; the top group is treated as Easy, the bottom group as Hard, and the intermediate group as Mid\. As shown in Figure[3](https://arxiv.org/html/2607.08662#Sx4.F3)\(a\) and Figure[3](https://arxiv.org/html/2607.08662#Sx4.F3)\(c\), WebSwarm’s advantage becomes more pronounced as task difficulty increases, especially on hard samples\. On the Hard subset, WebSwarm substantially improves over ReAct Agent: from 0\.0 to 35\.7 on BrowseComp\-Plus, and from 24\.5 to 55\.8 on WideSearch\-EN\. It also clearly outperforms other multi\-agent baselines\. Figures[3](https://arxiv.org/html/2607.08662#Sx4.F3)\(b\) and[3](https://arxiv.org/html/2607.08662#Sx4.F3)\(d\) show the corresponding web\-tool usage\. Multi\-agent methods generally make more tool calls than ReAct Agent\. WebSwarm adaptively increases resource use with task difficulty: it remains relatively controlled on Easy samples, while allocating substantially more search and reading budget to Hard samples\. These results indicate that WebSwarm matches other multi\-agent baselines on simple tasks while breaking the performance ceiling of ReAct Agent and multi\-agent baselines on complex and long\-tail samples\.

MethodBC\-PlusWideSearchDeepWideSearchACCRow F1Item F1Row F1Item F1Qwen3\-32BReAct12\.005\.4329\.113\.2820\.53WebSwarm19\.509\.0134\.477\.0625\.54Qwen3\.5\-35BReAct56\.5037\.3864\.8228\.7952\.46WebSwarm61\.0047\.5475\.9135\.9058\.34Table 4:Performance comparison across different models\.
### Performance across Different LLMs

To verify WebSwarm’s applicability across different backbones, we conduct experiments with Qwen3\-32B and Qwen3\.5\-35B, as shown in Table[4](https://arxiv.org/html/2607.08662#Sx4.T4)\. The ReAct baseline scores show that these backbones differ substantially in their native agentic search ability: Qwen3\-32B is relatively weak, while Qwen3\.5\-35B already exhibits strong long\-horizon web search capability\. Across both settings, WebSwarm consistently improves over ReAct on deep, wide, and deep\-wide search tasks\. These results indicate that WebSwarm’s gains are not tied to a specific backbone capability level\. For weaker models, recursive delegation provides additional structure for long\-horizon search; for stronger models, mode\-guided node solving, web\-structure\-guided expansion, and process\-level experience reuse further improve search depth and coverage\.

## Conclusion

In this paper, we propose WebSwarm for complex web information\-seeking tasks that require deep reasoning and broad information coverage\. Unlike static upfront decomposition or fixed collaboration patterns, WebSwarm formulates search as progressive recursive delegation over agentic search nodes\. Each node couples a local objective with a search mode, may solve locally or delegate child nodes, and returns evidence upward to support further expansion, revision, or aggregation\. By introducing Web\-structure\-guided expansion and experience reuse across homogeneous sibling nodes, WebSwarm jointly constructs task decomposition, recursive expansion, and agent collaboration as evidence accumulates\. Experiments across multiple benchmarks show that WebSwarm outperforms both single\-agent and multi\-agent baselines on deep, wide, and interleaved deep\-and\-wide tasks, with particularly clear gains on difficult examples\.

## References

- S\. Alzubi, B\. Nama, A\. Kaz, A\. Eswaran, W\. Chen, S\. Khetan, R\. Bala, T\. Vu, and S\. Oh \(2026a\)ROMA: recursive open meta\-agent framework for long\-horizon multi\-agent systems\.CoRRabs/2602\.01848\.External Links:[Link](https://doi.org/10.48550/arXiv.2602.01848),[Document](https://dx.doi.org/10.48550/ARXIV.2602.01848),2602\.01848Cited by:[6th item](https://arxiv.org/html/2607.08662#A3.I3.i6.p1.1),[Introduction](https://arxiv.org/html/2607.08662#Sx1.p2.1),[Multi\-Agent Systems for Long\-Horizon Tasks](https://arxiv.org/html/2607.08662#Sx2.SSx2.p1.1),[Baselines\.](https://arxiv.org/html/2607.08662#Sx4.SSx1.SSSx2.p1.1)\.
- S\. Alzubi, N\. Provenzano, J\. Bingham, W\. Chen, and T\. Vu \(2026b\)EvoSkill: automated skill discovery for multi\-agent systems\.CoRRabs/2603\.02766\.External Links:[Link](https://doi.org/10.48550/arXiv.2603.02766),[Document](https://dx.doi.org/10.48550/ARXIV.2603.02766),2603\.02766Cited by:[Experience\-Guided Node Solving\.](https://arxiv.org/html/2607.08662#Sx3.SSx3.SSSx2.p1.9)\.
- M\. Chen, G\. Zhang, H\. Chang, Y\. Guo, and S\. Zhou \(2026\)A\-mapreduce: executing wide search via agentic mapreduce\.CoRRabs/2602\.01331\.External Links:[Link](https://doi.org/10.48550/arXiv.2602.01331),[Document](https://dx.doi.org/10.48550/ARXIV.2602.01331),2602\.01331Cited by:[Introduction](https://arxiv.org/html/2607.08662#Sx1.p2.1),[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1),[Experience\-Guided Node Solving\.](https://arxiv.org/html/2607.08662#Sx3.SSx3.SSSx2.p1.9)\.
- Z\. Chen, K\. Liu, Q\. Wang, J\. Liu, W\. Zhang, K\. Chen, and F\. Zhao \(2025a\)MindSearch: mimicking human minds elicits deep AI searcher\.InThe Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24\-28, 2025,External Links:[Link](https://openreview.net/forum?id=xgtXkyqw1f)Cited by:[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- Z\. Chen, X\. Ma, S\. Zhuang, P\. Nie, K\. Zou, A\. Liu, J\. Green, K\. Patel, R\. Meng, M\. Su, S\. Sharifymoghaddam, Y\. Li, H\. Hong, X\. Shi, X\. Liu, N\. Thakur, C\. Zhang, L\. Gao, W\. Chen, and J\. Lin \(2025b\)BrowseComp\-plus: A more fair and transparent evaluation benchmark of deep\-research agent\.CoRRabs/2508\.06600\.External Links:[Link](https://doi.org/10.48550/arXiv.2508.06600),[Document](https://dx.doi.org/10.48550/ARXIV.2508.06600),2508\.06600Cited by:[1st item](https://arxiv.org/html/2607.08662#A3.I2.i1.p1.1),[Introduction](https://arxiv.org/html/2607.08662#Sx1.p1.1),[Benchmarks\.](https://arxiv.org/html/2607.08662#Sx4.SSx1.SSSx1.p1.1)\.
- A\. Fourney, G\. Bansal, H\. Mozannar, C\. Tan, E\. Salinas, E\. Zhu, F\. Niedtner, G\. Proebsting, G\. Bassman, J\. Gerrits, J\. Alber, P\. Chang, R\. Loynd, R\. West, V\. Dibia, A\. Awadallah, E\. Kamar, R\. Hosn, and S\. Amershi \(2024\)Magentic\-one: A generalist multi\-agent system for solving complex tasks\.CoRRabs/2411\.04468\.External Links:[Link](https://doi.org/10.48550/arXiv.2411.04468),[Document](https://dx.doi.org/10.48550/ARXIV.2411.04468),2411\.04468Cited by:[Multi\-Agent Systems for Long\-Horizon Tasks](https://arxiv.org/html/2607.08662#Sx2.SSx2.p1.1)\.
- S\. Hong, M\. Zhuge, J\. Chen, X\. Zheng, Y\. Cheng, J\. Wang, C\. Zhang, Z\. Wang, S\. K\. S\. Yau, Z\. Lin, L\. Zhou, C\. Ran, L\. Xiao, C\. Wu, and J\. Schmidhuber \(2024\)MetaGPT: meta programming for A multi\-agent collaborative framework\.InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7\-11, 2024,External Links:[Link](https://openreview.net/forum?id=VtmBAGCN7o)Cited by:[Multi\-Agent Systems for Long\-Horizon Tasks](https://arxiv.org/html/2607.08662#Sx2.SSx2.p1.1)\.
- Y\. Hu, H\. Qian, S\. Wang, J\. Liu, T\. Zhao, X\. Li, Z\. Liu, and Z\. Dou \(2026\)AgentFugue: agent scaling for long\-horizon tasks through collective reasoning\.CoRRabs/2605\.24486\.External Links:[Link](https://doi.org/10.48550/arXiv.2605.24486),[Document](https://dx.doi.org/10.48550/ARXIV.2605.24486),2605\.24486Cited by:[Multi\-Agent Systems for Long\-Horizon Tasks](https://arxiv.org/html/2607.08662#Sx2.SSx2.p1.1)\.
- Y\. Huang, Y\. Chen, Z\. He, Y\. Chen, K\. Y\. Lee, H\. Zhou, W\. Luo, M\. Fang, and J\. Wang \(2026\)Web2BigTable: A bi\-level multi\-agent LLM system for internet\-scale information search and extraction\.CoRRabs/2604\.27221\.External Links:[Link](https://doi.org/10.48550/arXiv.2604.27221),[Document](https://dx.doi.org/10.48550/ARXIV.2604.27221),2604\.27221Cited by:[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- B\. Jin, H\. Zeng, Z\. Yue, J\. Yoon, S\. O\. Arik, D\. Wang, H\. Zamani, and J\. Han \(2025a\)Search\-r1: training LLMs to reason and leverage search engines with reinforcement learning\.InSecond Conference on Language Modeling,External Links:[Link](https://openreview.net/forum?id=Rwhi91ideu)Cited by:[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- J\. Jin, X\. Li, G\. Dong, Y\. Zhang, Y\. Zhu, Z\. Yang, H\. Qian, and Z\. Dou \(2025b\)Decoupled planning and execution: A hierarchical reasoning framework for deep search\.CoRRabs/2507\.02652\.External Links:[Link](https://doi.org/10.48550/arXiv.2507.02652),[Document](https://dx.doi.org/10.48550/ARXIV.2507.02652),2507\.02652Cited by:[Introduction](https://arxiv.org/html/2607.08662#Sx1.p2.1),[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- T\. Lan, F\. Henry, B\. Zhu, Q\. Jia, J\. Ren, Q\. Pu, H\. Li, L\. Wang, Z\. Xu, and W\. Luo \(2026\)Table\-as\-search: formulate long\-horizon agentic information seeking as table completion\.CoRRabs/2602\.06724\.External Links:[Link](https://doi.org/10.48550/arXiv.2602.06724),[Document](https://dx.doi.org/10.48550/ARXIV.2602.06724),2602\.06724Cited by:[4th item](https://arxiv.org/html/2607.08662#A3.I3.i4.p1.1),[Introduction](https://arxiv.org/html/2607.08662#Sx1.p2.1),[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1),[Baselines\.](https://arxiv.org/html/2607.08662#Sx4.SSx1.SSSx2.p1.1)\.
- T\. Lan, B\. Zhu, Q\. Jia, J\. Ren, H\. Li, L\. Wang, Z\. Xu, W\. Luo, and K\. Zhang \(2025\)DeepWideSearch: benchmarking depth and width in agentic information seeking\.CoRRabs/2510\.20168\.External Links:[Link](https://doi.org/10.48550/arXiv.2510.20168),[Document](https://dx.doi.org/10.48550/ARXIV.2510.20168),2510\.20168Cited by:[3rd item](https://arxiv.org/html/2607.08662#A3.I2.i3.p1.1),[Introduction](https://arxiv.org/html/2607.08662#Sx1.p1.1),[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1),[Benchmarks\.](https://arxiv.org/html/2607.08662#Sx4.SSx1.SSSx1.p1.1)\.
- K\. Y\. Lee, Y\. Huang, Z\. He, H\. Zhou, W\. Luo, K\. Shao, M\. Fang, and J\. Wang \(2026a\)InfoSeeker: A scalable hierarchical parallel agent framework for web information seeking\.CoRRabs/2604\.02971\.External Links:[Link](https://doi.org/10.48550/arXiv.2604.02971),[Document](https://dx.doi.org/10.48550/ARXIV.2604.02971),2604\.02971Cited by:[5th item](https://arxiv.org/html/2607.08662#A3.I3.i5.p1.1),[Introduction](https://arxiv.org/html/2607.08662#Sx1.p2.1),[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1),[Baselines\.](https://arxiv.org/html/2607.08662#Sx4.SSx1.SSSx2.p1.1)\.
- Y\. Lee, H\. Yen, X\. Ye, and D\. Chen \(2026b\)Agentic aggregation for parallel scaling of long\-horizon agentic tasks\.CoRRabs/2604\.11753\.External Links:[Link](https://doi.org/10.48550/arXiv.2604.11753),[Document](https://dx.doi.org/10.48550/ARXIV.2604.11753),2604\.11753Cited by:[Multi\-Agent Systems for Long\-Horizon Tasks](https://arxiv.org/html/2607.08662#Sx2.SSx2.p1.1)\.
- B\. Li, B\. Zhang, D\. Zhang, F\. Huang, G\. Li, G\. Chen, H\. Yin, J\. Wu, J\. Zhou, K\. Li, L\. Su, L\. Ou, L\. Zhang, P\. Xie, R\. Ye, W\. Yin, X\. Yu, X\. Wang, X\. Wu, X\. Chen, Y\. Zhao, Z\. Zhang, Z\. Tao, Z\. Zhang, Z\. Qiao, C\. Wang, D\. Yu, G\. Fu, H\. Shen, J\. Yang, J\. Lin, J\. Zhang, K\. Zeng, L\. Yang, H\. Yin, M\. Song, M\. Yan, P\. Xia, Q\. Xiao, R\. Min, R\. Ding, R\. Fang, S\. Chen, S\. Huang, S\. Wang, S\. Cai, W\. Shen, X\. Wang, X\. Guan, X\. Geng, Y\. Shi, Y\. Wu, Z\. Chen, Z\. Li, and Y\. Jiang \(2025a\)Tongyi deepresearch technical report\.CoRRabs/2510\.24701\.External Links:[Link](https://doi.org/10.48550/arXiv.2510.24701),[Document](https://dx.doi.org/10.48550/ARXIV.2510.24701),2510\.24701Cited by:[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- G\. Li, H\. Hammoud, H\. Itani, D\. Khizbullin, and B\. Ghanem \(2023\)CAMEL: communicative agents for "mind" exploration of large language model society\.InAdvances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, New Orleans, LA, USA, December 10 \- 16, 2023,A\. Oh, T\. Naumann, A\. Globerson, K\. Saenko, M\. Hardt, and S\. Levine \(Eds\.\),External Links:[Link](http://papers.nips.cc/paper%5C_files/paper/2023/hash/a3621ee907def47c1b952ade25c67698-Abstract-Conference.html)Cited by:[Multi\-Agent Systems for Long\-Horizon Tasks](https://arxiv.org/html/2607.08662#Sx2.SSx2.p1.1)\.
- X\. Li, G\. Dong, J\. Jin, Y\. Zhang, Y\. Zhou, Y\. Zhu, P\. Zhang, and Z\. Dou \(2025b\)Search\-o1: agentic search\-enhanced large reasoning models\.InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, EMNLP 2025, Suzhou, China, November 4\-9, 2025,C\. Christodoulopoulos, T\. Chakraborty, C\. Rose, and V\. Peng \(Eds\.\),pp\. 5420–5438\.External Links:[Link](https://doi.org/10.18653/v1/2025.emnlp-main.276),[Document](https://dx.doi.org/10.18653/V1/2025.EMNLP-MAIN.276)Cited by:[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- X\. Li, J\. Jin, G\. Dong, H\. Qian, Y\. Wu, J\. Wen, Y\. Zhu, and Z\. Dou \(2026\)Webthinker: empowering large reasoning models with deep research capability\.Advances in Neural Information Processing Systems38,pp\. 120091–120131\.Cited by:[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- J\. Liu, X\. Zhao, X\. Shang, and Z\. Shen \(2026\)Dive into claude code: the design space of today’s and future AI agent systems\.CoRRabs/2604\.14228\.External Links:[Link](https://doi.org/10.48550/arXiv.2604.14228),[Document](https://dx.doi.org/10.48550/ARXIV.2604.14228),2604\.14228Cited by:[Multi\-Agent Systems for Long\-Horizon Tasks](https://arxiv.org/html/2607.08662#Sx2.SSx2.p1.1)\.
- R\. Nakano, J\. Hilton, S\. Balaji, J\. Wu, L\. Ouyang, C\. Kim, C\. Hesse, S\. Jain, V\. Kosaraju, W\. Saunders, X\. Jiang, K\. Cobbe, T\. Eloundou, G\. Krueger, K\. Button, M\. Knight, B\. Chess, and J\. Schulman \(2021\)WebGPT: browser\-assisted question\-answering with human feedback\.CoRRabs/2112\.09332\.External Links:[Link](https://arxiv.org/abs/2112.09332),2112\.09332Cited by:[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- P\. Ning, Q\. Chen, K\. Tao, X\. Tang, T\. Wang, Q\. Cao, X\. Kong, Z\. Wen, Z\. Zhang, and J\. Zhou \(2026\)SearchSwarm: towards delegation intelligence in agentic llms for long\-horizon deep research\.External Links:2606\.09730,[Link](https://arxiv.org/abs/2606.09730)Cited by:[Introduction](https://arxiv.org/html/2607.08662#Sx1.p2.1),[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- C\. Qian, W\. Liu, H\. Liu, N\. Chen, Y\. Dang, J\. Li, C\. Yang, W\. Chen, Y\. Su, X\. Cong, J\. Xu, D\. Li, Z\. Liu, and M\. Sun \(2024\)ChatDev: communicative agents for software development\.InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics \(Volume 1: Long Papers\), ACL 2024, Bangkok, Thailand, August 11\-16, 2024,L\. Ku, A\. Martins, and V\. Srikumar \(Eds\.\),pp\. 15174–15186\.External Links:[Link](https://doi.org/10.18653/v1/2024.acl-long.810),[Document](https://dx.doi.org/10.18653/V1/2024.ACL-LONG.810)Cited by:[Multi\-Agent Systems for Long\-Horizon Tasks](https://arxiv.org/html/2607.08662#Sx2.SSx2.p1.1)\.
- T\. Qin, Q\. Chen, S\. Wang, H\. Xing, K\. Zhu, H\. Zhu, D\. Shi, X\. Liu, G\. Zhang, J\. Liu, X\. Gao, Y\. E\. Jiang, and W\. Zhou \(2026\)Flash\-searcher: fast and effective web agents via DAG\-based parallel execution\.InThe Fourteenth International Conference on Learning Representations,External Links:[Link](https://openreview.net/forum?id=QuaJ6kJaBm)Cited by:[3rd item](https://arxiv.org/html/2607.08662#A3.I3.i3.p1.1),[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1),[Baselines\.](https://arxiv.org/html/2607.08662#Sx4.SSx1.SSSx2.p1.1)\.
- K\. Team \(2026\)Kimi K2\.5: visual agentic intelligence\.CoRRabs/2602\.02276\.External Links:[Link](https://doi.org/10.48550/arXiv.2602.02276),[Document](https://dx.doi.org/10.48550/ARXIV.2602.02276),2602\.02276Cited by:[2nd item](https://arxiv.org/html/2607.08662#A3.I3.i2.p1.1),[Introduction](https://arxiv.org/html/2607.08662#Sx1.p2.1),[Multi\-Agent Systems for Long\-Horizon Tasks](https://arxiv.org/html/2607.08662#Sx2.SSx2.p1.1),[Baselines\.](https://arxiv.org/html/2607.08662#Sx4.SSx1.SSSx2.p1.1)\.
- J\. Wei, Z\. Sun, S\. Papay, S\. McKinney, J\. Han, I\. Fulford, H\. W\. Chung, A\. T\. Passos, W\. Fedus, and A\. Glaese \(2025\)BrowseComp: A simple yet challenging benchmark for browsing agents\.CoRRabs/2504\.12516\.External Links:[Link](https://doi.org/10.48550/arXiv.2504.12516),[Document](https://dx.doi.org/10.48550/ARXIV.2504.12516),2504\.12516Cited by:[1st item](https://arxiv.org/html/2607.08662#A3.I2.i1.p1.1),[Introduction](https://arxiv.org/html/2607.08662#Sx1.p1.1),[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- R\. Wong, J\. Wang, J\. zhao, L\. Chen, Y\. Gao, Zhanglong, X\. Zhou, Z\. Wang, K\. Xiang, G\. Zhang, W\. Huang, Y\. Wang, and W\. KE \(2026\)WideSearch: benchmarking agentic broad info\-seeking\.InThe Fourteenth International Conference on Learning Representations,External Links:[Link](https://openreview.net/forum?id=Q7YUY7zGkZ)Cited by:[2nd item](https://arxiv.org/html/2607.08662#A3.I2.i2.p1.1),[Introduction](https://arxiv.org/html/2607.08662#Sx1.p1.1),[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1),[Benchmarks\.](https://arxiv.org/html/2607.08662#Sx4.SSx1.SSSx1.p1.1)\.
- J\. Wu, B\. Li, R\. Fang, W\. Yin, L\. Zhang, Z\. Wang, Z\. Tao, D\. Zhang, Z\. Xi, R\. Tang,et al\.\(2026\)Webdancer: towards autonomous information seeking agency\.Advances in Neural Information Processing Systems38,pp\. 120957–120985\.Cited by:[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- Q\. Wu, G\. Bansal, J\. Zhang, Y\. Wu, S\. Zhang, E\. Zhu, B\. Li, L\. Jiang, X\. Zhang, and C\. Wang \(2023\)AutoGen: enabling next\-gen LLM applications via multi\-agent conversation framework\.CoRRabs/2308\.08155\.External Links:[Link](https://doi.org/10.48550/arXiv.2308.08155),[Document](https://dx.doi.org/10.48550/ARXIV.2308.08155),2308\.08155Cited by:[Multi\-Agent Systems for Long\-Horizon Tasks](https://arxiv.org/html/2607.08662#Sx2.SSx2.p1.1)\.
- Y\. Xi, J\. Lin, Y\. Xiao, Z\. Zhou, R\. Shan, T\. Gao, J\. Zhu, W\. Liu, Y\. Yu, and W\. Zhang \(2025\)A survey of llm\-based deep search agents: paradigm, optimization, evaluation, and challenges\.CoRRabs/2508\.05668\.External Links:[Link](https://doi.org/10.48550/arXiv.2508.05668),[Document](https://dx.doi.org/10.48550/ARXIV.2508.05668),2508\.05668Cited by:[Introduction](https://arxiv.org/html/2607.08662#Sx1.p1.1),[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- Z\. Xu, Z\. Xu, R\. Zhang, C\. Zhu, S\. Yu, W\. Liu, Q\. Zhang, W\. Ding, C\. Yu, and Y\. Wang \(2026\)WideSeek\-r1: exploring width scaling for broad information seeking via multi\-agent reinforcement learning\.CoRRabs/2602\.04634\.External Links:[Link](https://doi.org/10.48550/arXiv.2602.04634),[Document](https://dx.doi.org/10.48550/ARXIV.2602.04634),2602\.04634Cited by:[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- S\. Yao, J\. Zhao, D\. Yu, N\. Du, I\. Shafran, K\. R\. Narasimhan, and Y\. Cao \(2023\)ReAct: synergizing reasoning and acting in language models\.InThe Eleventh International Conference on Learning Representations, ICLR 2023, Kigali, Rwanda, May 1\-5, 2023,External Links:[Link](https://openreview.net/forum?id=WE%5C_vluYUL-X)Cited by:[1st item](https://arxiv.org/html/2607.08662#A3.I3.i1.p1.1),[Introduction](https://arxiv.org/html/2607.08662#Sx1.p2.1),[Baselines\.](https://arxiv.org/html/2607.08662#Sx4.SSx1.SSSx2.p1.1)\.
- Y\. Zhang, M\. Li, D\. Long, X\. Zhang, H\. Lin, B\. Yang, P\. Xie, A\. Yang, D\. Liu, J\. Lin, F\. Huang, and J\. Zhou \(2025\)Qwen3 embedding: advancing text embedding and reranking through foundation models\.CoRRabs/2506\.05176\.External Links:[Link](https://doi.org/10.48550/arXiv.2506.05176),[Document](https://dx.doi.org/10.48550/ARXIV.2506.05176),2506\.05176Cited by:[1st item](https://arxiv.org/html/2607.08662#A3.I2.i1.p1.1)\.
- Y\. Zhu, H\. Yuan, S\. Wang, J\. Liu, W\. Liu, C\. Deng, H\. Chen, Z\. Liu, Z\. Dou, and J\. Wen \(2026a\)Large language models for information retrieval: A survey\.ACM Trans\. Inf\. Syst\.44\(1\),pp\. 12:1–12:54\.External Links:[Link](https://doi.org/10.1145/3748304),[Document](https://dx.doi.org/10.1145/3748304)Cited by:[Introduction](https://arxiv.org/html/2607.08662#Sx1.p1.1),[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1)\.
- Y\. Zhu, X\. Zhang, M\. Zhang, J\. Jin, L\. Zhang, X\. Song, K\. Zhao, W\. Zeng, R\. Tang, H\. Li, J\. Wen, and Z\. Dou \(2026b\)GISA: A benchmark for general information\-seeking assistant\.CoRRabs/2602\.08543\.External Links:[Link](https://doi.org/10.48550/arXiv.2602.08543),[Document](https://dx.doi.org/10.48550/ARXIV.2602.08543),2602\.08543Cited by:[4th item](https://arxiv.org/html/2607.08662#A3.I2.i4.p1.1),[Introduction](https://arxiv.org/html/2607.08662#Sx1.p1.1),[Agent for Web Information Seeking](https://arxiv.org/html/2607.08662#Sx2.SSx1.p1.1),[Benchmarks\.](https://arxiv.org/html/2607.08662#Sx4.SSx1.SSSx1.p1.1)\.

## Appendix ALimitations

In this paper, we propose WebSwarm, a progressive recursive delegation framework for complex web information\-seeking tasks\. Although WebSwarm improves both deep and wide search, it still has several limitations\. \(1\) In terms of research scope, WebSwarm mainly focuses on inference\-time multi\-agent orchestration, without involving data construction or policy optimization for multi\-agent training\. Therefore, node delegation, search\-mode assignment, Web\-structure\-guided expansion, and subtask experience reuse still rely on the understanding and reasoning abilities of the base LLM, which may not fully unlock the model’s potential for recursive multi\-agent collaboration\. \(2\) In terms of resource consumption, although WebSwarm adaptively instantiates search nodes and expands the delegation structure according to task difficulty, it still requires more LLM calls and web\-tool requests than a single ReAct agent, leading to higher inference cost and latency\. \(3\) In terms of modality, WebSwarm mainly targets information\-seeking tasks based on text\-based web tools and has not yet fully covered multimodal web information, such as images, videos, and audio\. Looking to the future, WebSwarm can be extended in two directions\. First, multi\-agent training can enable models to learn better recursive delegation, search\-mode assignment, and collaboration strategies from search trajectories\. Second, the current text\-based web\-tool framework can be expanded to multimodal and GUI\-based web scenarios\.

## Appendix BDetails of WebSwarm

### Details of Prompts

Figure[8](https://arxiv.org/html/2607.08662#A4.F8)shows the system prompt of the root agent, which instructs it to iteratively generate subtasks and select their execution types\. Figure[9](https://arxiv.org/html/2607.08662#A4.F9)presents the system prompt of the web structure probing agent, which guides the agent to pre\-interact with the web and understand how task\-relevant information is organized online\. Figure[10](https://arxiv.org/html/2607.08662#A4.F10)shows the system prompt for subtask\-skill summarization, which instructs the LLM to summarize reusable experience from completed subtask trajectories for subsequent subtasks\.

### Pseudo\-code

We summarize the pseudo\-code of WebSwarm inference in Algorithm 1\.

Algorithm 1WebSwarm Inference1:user task

q0q\_\{0\}
2:final answer

aa
3:

qvq\_\{v\}: local objective;

mvm\_\{v\}: search mode;

RvR\_\{v\}: child results;

hvh\_\{v\}: web\-structure hint;

CvC\_\{v\}: child delegations;

kk: injected experience;

τv\\tau\_\{v\}: solving trajectory\.

4:

v0←CreateNode​\(q0\)v\_\{0\}\\leftarrow\\textsc\{CreateNode\}\(q\_\{0\}\)
5:

\(a,τv0\)←SolveNode​\(v0,∅\)\(a,\\tau\_\{v\_\{0\}\}\)\\leftarrow\\textsc\{SolveNode\}\(v\_\{0\},\\emptyset\)
6:return

aa
7:functionSolveNode\(

v,kv,k\)

8:

Rv,τv←∅,∅R\_\{v\},\\tau\_\{v\}\\leftarrow\\emptyset,\\emptyset
9:whilenot

Ready​\(qv,Rv\)\\textsc\{Ready\}\(q\_\{v\},R\_\{v\}\)do

10:

hv←WebProbing​\(qv\)h\_\{v\}\\leftarrow\\textsc\{WebProbing\}\(q\_\{v\}\)if

NeedProbe​\(v,Rv\)\\textsc\{NeedProbe\}\(v,R\_\{v\}\)else

∅\\emptyset
11:

\(y,τy\)←SolveOrDelegate​\(qv,mv,Rv,hv,k\)\(y,\\tau\_\{y\}\)\\leftarrow\\textsc\{SolveOrDelegate\}\(q\_\{v\},m\_\{v\},R\_\{v\},h\_\{v\},k\)
12:

τv←τv∘τy\\tau\_\{v\}\\leftarrow\\tau\_\{v\}\\circ\\tau\_\{y\}
13:if

yyis a resultthen

14:return

\(y,τv\)\(y,\\tau\_\{v\}\)
15:endif

16:

\(Cv,ks,Rs,Ts\)←ScoutExperience​\(v,y\)\(C\_\{v\},k^\{s\},R^\{s\},T^\{s\}\)\\leftarrow\\textsc\{ScoutExperience\}\(v,y\)
17:

\(Rc,Tc\)←RunChildren​\(v,Cv,ks\)\(R^\{c\},T^\{c\}\)\\leftarrow\\textsc\{RunChildren\}\(v,C\_\{v\},k^\{s\}\)
18:

Rv←Rv∪Rs∪RcR\_\{v\}\\leftarrow R\_\{v\}\\cup R^\{s\}\\cup R^\{c\}
19:endwhile

20:return

\(Aggregate​\(qv,Rv\),τv\)\(\\textsc\{Aggregate\}\(q\_\{v\},R\_\{v\}\),\\tau\_\{v\}\)
21:endfunction

22:functionScoutExperience\(

v,Cv,C\)

23:ifnot

NeedExperience​\(C\)\\textsc\{NeedExperience\}\(C\)then

24:return

\(C,∅,∅,∅\)\(C,\\emptyset,\\emptyset,\\emptyset\)
25:endif

26:

Cs←SelectScouts​\(C\)C^\{s\}\\leftarrow\\textsc\{SelectScouts\}\(C\)
27:

\(Rs,Ts\)←RunChildren​\(v,Cs,∅\)\(R^\{s\},T^\{s\}\)\\leftarrow\\textsc\{RunChildren\}\(v,C^\{s\},\\emptyset\)
28:return

\(C∖Cs,ExtractExperience​\(Ts\),Rs,Ts\)\(C\\setminus C^\{s\},\\textsc\{ExtractExperience\}\(T^\{s\}\),R^\{s\},T^\{s\}\)
29:endfunction

30:functionRunChildren\(

v,C,kv,C,k\)

31:

R,T←∅,∅R,T\\leftarrow\\emptyset,\\emptyset
32:for all

\(qi,mi\)∈C\(q\_\{i\},m\_\{i\}\)\\in Cdo

33:

ui←CreateChild​\(v,qi,mi\)u\_\{i\}\\leftarrow\\textsc\{CreateChild\}\(v,q\_\{i\},m\_\{i\}\)
34:

\(ri,τi\)←SolveNode​\(ui,k\)\(r\_\{i\},\\tau\_\{i\}\)\\leftarrow\\textsc\{SolveNode\}\(u\_\{i\},k\)
35:

R←R∪\{ri\},T←T∘τiR\\leftarrow R\\cup\\\{r\_\{i\}\\\},\\quad T\\leftarrow T\\circ\\tau\_\{i\}
36:endfor

37:return

\(R,T\)\(R,T\)
38:endfunction

## Appendix CDetails of Experiments

### Details of Web Tools

WebSwarm and all baselines interact with the web through a unified tool interface:

- •Web\-Search Tool: Given a search query, the Web\-Search Tool calls the Serper API222https://serper\.dev/to retrieve candidate web pages\. The returned results are converted into a structured format containing the title, URL, snippet, date, and other available metadata\. To control the search space, the tool returns the top\-5 search results for each query\.
- •Page\-Browse Tool: Given a URL, the Page\-Browse Tool uses the Jina Reader API333https://jina\.ai/readerto fetch the page content in Markdown format, with the page length limited to 80K tokens\. Since raw web pages can be long and noisy, we further use the same LLM as the experimental agent to generate a goal\-conditioned summary of the page\. The summarized content is then returned to the agent for evidence verification, table filling, or subsequent reasoning\.

Figure[11](https://arxiv.org/html/2607.08662#A4.F11)shows the tool descriptions provided to agents\.

### Details of Agent Inference

For all locally deployed models running on NVIDIA A800 GPUs, we use SGLang444https://github\.com/sgl\-project/sglangas the inference engine and keep the same hyperparameters throughout the experiments: thinking mode is enabled, temperature is set to0\.60\.6, top\-ppto0\.950\.95, top\-kkto2020, repetition penalty to1\.01\.0, presence penalty to1\.51\.5, and the maximum number of generated tokens to3276832768\. To avoid prematurely truncating agent actions, we use the maximum context length supported by each model, namely 128K tokens for GLM\-4\.5 and 256K tokens for Qwen3\.5, and allow each agent to execute up to 200 action steps\. We run the experiments twice randomly and take the average\.

### Details of Evaluation Benchmarks

- •BrowseComp\-Plus\(Chenet al\.[2025b](https://arxiv.org/html/2607.08662#bib.bib4)\): BrowseComp\-Plus shares the same tasks as BrowseComp\(Weiet al\.[2025](https://arxiv.org/html/2607.08662#bib.bib3)\), but implements it through retrieval over a fixed local document corpus instead of live web search, enabling more reproducible comparisons across agent frameworks\. Following the original benchmark setting, we use Qwen3\-Embedding\-8B\(Zhanget al\.[2025](https://arxiv.org/html/2607.08662#bib.bib37)\)as the retrieval model\. Its key metric is answer\-equivalence accuracy, which measures whether the predicted answer is semantically equivalent to the reference answer\. In our experiments, we randomly sample 200 tasks for evaluation\.
- •WideSearch\(Wonget al\.[2026](https://arxiv.org/html/2607.08662#bib.bib2)\): WideSearch focuses on broad information collection, where a system needs to identify a set of relevant entities and fill their attributes into a structured table\. Its main metrics include Item\-F1, Row\-F1, and Success Rate: Item\-F1 measures cell\-level correctness, Row\-F1 measures whether complete entity records are recovered, and Success Rate requires the whole table to match the reference answer\. The benchmark contains both Chinese and English tasks; in our experiments, we evaluate only the English subset, consisting of 100 tasks, to reduce resource costs\.
- •DeepWideSearch\(Lanet al\.[2025](https://arxiv.org/html/2607.08662#bib.bib5)\): DeepWideSearch covers tasks that require both deep search and broad information coverage\. It is constructed in two ways: extending deep\-search tasks into broad table\-based information collection, and augmenting broad\-search tasks with prerequisite deep\-search steps\. Its evaluation includes intermediate metrics, namely Column\-F1 and Core Entity Accuracy, as well as final table\-level metrics, including Item\-F1, Row\-F1, and Success Rate\. Since the intermediate metrics in this benchmark are highly unreliable, we only report the final table\-level metrics\. The dataset contains both Chinese and English tasks; in our experiments, we evaluate only the English subset, consisting of 76 tasks, to reduce resource costs\.
- •GISA\(Zhuet al\.[2026b](https://arxiv.org/html/2607.08662#bib.bib1)\): GISA is a general information\-seeking benchmark covering multiple answer formats, including item, set, list, and table\. Each format has corresponding metrics: Exact Match \(EM\) evaluates exact answer correctness, Set\-F1 measures unordered set coverage, List\-Content\-F1 and List\-Order\-Score evaluate list completeness and ordering, and Table\-Item\-F1 and Table\-Row\-F1 evaluate structured table answers at the cell and row levels\. This benchmark is entirely in English; in our experiments, we evaluate all 373 tasks\.

### Details of Baselines

- •ReAct Agent\(Yaoet al\.[2023](https://arxiv.org/html/2607.08662#bib.bib16)\): The ReAct Agent uses one agent to complete the entire task through interleaved reasoning and tool actions\. It performs search, page reading, evidence analysis, and answer generation within a single context, without explicitly creating subagents or maintaining a structured task decomposition\.
- •Swarm\-Agent\(Team[2026](https://arxiv.org/html/2607.08662#bib.bib7)\): Swarm\-Agent follows the inference framework of Kimi\-Swarm\(Team[2026](https://arxiv.org/html/2607.08662#bib.bib7)\), where a coordinator can dynamically create subagents and assign subtasks to them\. Concretely, the coordinator uses tools includingcreate\_subagentsto instantiate role\-specific agents andassign\_tasksto delegate subtasks; the subagents then return their intermediate results to the coordinator for further reasoning, aggregation, or additional decomposition\.
- •Flash\-Searcher\(Qinet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib8)\): Flash\-Searcher organizes the search process as a dependency\-aware directed acyclic graph \(DAG\) rather than a single linear trajectory\. It decomposes a complex query into subtasks with explicit dependencies, tracks the logical constraints among them, and executes independent reasoning branches in parallel\. The workflow can be dynamically optimized based on intermediate results, enabling more efficient tool use and reducing redundant sequential search steps\.
- •Table\-as\-Search\(Lanet al\.[2026](https://arxiv.org/html/2607.08662#bib.bib9)\): Table\-as\-Search represents long\-horizon information seeking as a table\-completion process\. The external table acts as an explicit search state: rows usually correspond to candidate entities, columns correspond to constraints or target attributes, filled cells store verified information, and empty cells indicate remaining search targets\.
- •InfoSeeker\(Leeet al\.[2026a](https://arxiv.org/html/2607.08662#bib.bib12)\): InfoSeeker adopts a hierarchical Host–Manager–Worker architecture\. The Host performs high\-level planning, Managers decompose and coordinate domain\-specific subtasks, and Workers execute atomic subtasks\. For a fair comparison, we replace the search APIs used in the original implementation with the same Serper\-based search tool and Jina\-based page\-reading tool used in our experiments, while keeping the tool configuration consistent with other methods\.
- •ROMA\(Alzubiet al\.[2026a](https://arxiv.org/html/2607.08662#bib.bib13)\): ROMA is a recursive meta\-agent framework that decomposes complex tasks into smaller subtasks and aggregates their results hierarchically\. Its workflow typically consists of atomizing a task, planning dependency\-aware subtasks, executing atomic subtasks, and aggregating the intermediate outputs through recursive control\.

## Appendix DCase Study

To better illustrate how WebSwarm works across different complex search tasks, we provide several representative case studies as follows:

- •Figure[4](https://arxiv.org/html/2607.08662#A4.F4)\(BrowseComp\-Plus\)\.This BrowseComp\-Plus case requires identifying an unknown entity from multiple indirect constraints\. WebSwarm instantiates adeepsearch node, where multiple Searchers explore different clue paths and a Verifier checks the candidate against all constraints\. The system finally identifies Walter Allen, demonstrating the candidate generation, verification, and refinement process for deep search\.
- •Figure[5](https://arxiv.org/html/2607.08662#A4.F5)\(WideSearch\)\.This case asks the system to build a product table across multiple spirits brands\. Since the information is organized by brand on the web, WebSwarm expands into brand\-level child nodes, first usingentity\_collectto obtain each brand’s product set and then delegatingatomsearch nodes to fill product attributes\. This illustrates recursive wide search following the structure “brand→\\rightarrowproduct→\\rightarrowattribute\.”
- •Figure[6](https://arxiv.org/html/2607.08662#A4.F6)\(WideSearch\)\.This case requires extracting Taylor Swift’s official concert records over a long time range\. web\-structure probing indicates that the data is mainly organized by tour rather than by year, so WebSwarm expands along official tours and delegates parallelatomsearch nodes to extract show\-level rows\. This shows how web\-structure\-guided expansion helps choose a more effective expansion axis\.
- •Figure[7](https://arxiv.org/html/2607.08662#A4.F7)\(DeepWideSearch\)\.This case first requires resolving a hidden manufacturer clue and then collecting structured vehicle information\. WebSwarm first uses adeepsearch node to identify Henry Ford, then expands into awidenode that enumerates eligible Ford models withentity\_collectand fills attributes withatomnodes\. This demonstrates an evidence\-driven transition from deep entity identification to wide structured collection\.

Original QuestionTwo people owned a business in a suburban town in the northwestern United States after 1900\. The primary activity of the business focused on people and the surrounding area\. The business moved locations to a street named for a number between 1 and 10\. After 1910 but before 1920 one of the owners was no longer listed\. The remaining owner worked with his wife and the business changed locations two more times\. The wife’s initials were G\.F\.P\. What was the name of the owner who was no longer listed?

Recursive Deep Search DelegationRoot Node Receive the original question and initiate recursive delegation\.Delegate Subtask 1 Find a business owned by two people in a suburban town in the northwestern US after 1900\. The business focused on people and the surrounding area \(likely a service business\)\. It moved to a street named for a number between 1 and 10\. After 1910 but before 1920, one of the original owners was no longer listed as an owner\. The remaining owner continued working with his wife, whose initials were G\.F\.P\. The business changed locations two more times\. What was the name of the owner who was no longer listed?Deep Node ReceiveSubtask 1, organize iterative searcher–verifier through child search nodes\.Searcher 1 Task: two people northwestern US numbered street 1910–1920; wife initials G\.F\.P\.Output: Unknown\.Searcher 2 Task: wife initials G\.F\.P\.; northwestern US 1910s service company partnership\.Output: Unknown\.Searcher 3 Task: business moved First/Second/Third Street; partnership dissolved 1910–1920; spouse\.Output: Unknown\.Searcher 4 Task: photography studio partnership dissolved 1910–1920; northwestern United States; wife on numbered street\.Output: Found candidate: the photography studio partnership of Frank Perkins and Walter Allen was dissolved around 1916\-1917…Searcher 5 Task: Frank Perkins Walter Allen Georgetown Photograph Studio moved locations after 1916\.Output: Found supporting move history\. \(Frank Perkins Walter Allen’s Georgetown Photograph Studio in Seattle moved locations multiple times after 1916…\)Verifier Check whether Walter Allen satisfies all constraints: a two\-owner business after 1900; relocation to a numbered street; one owner no longer listed between 1910 and 1920…Output:✓Verified: all constraints match\.Deep Node Result Walter Allen✓Root Node completion Receive the returned result and determine whether the original task is solved\.Answer obtained\. Task complete\.✓

Figure 4:Case study for bc\_plus\_92 in the BrowseComp\-Plus benchmark\.Original QuestionI’m currently mapping the product portfolios of several spirits brands including Johnnie Walker, Chivas Regal, Smirnoff, Grey Goose, Absolut Vodka, and Bacardi as of June 2025\. The scope covers only their standard products in the Core / Permanent Range, excluding any flavor variants or limited\-edition / seasonal releases\. Please organize the results in one Markdown table with the columns: Brand, Product, Category, Sub\-category, Pack Size \(Bottle\), ABV %\. For any missing information, fill in ‘/‘\. Please ensure that no cells in the Brand column are left blank\. Do not ask questions; directly output the results in the required table format\.

Recursive Wide Search DelegationRoot Nodereceive the request, initialize recursive delegation, and create a top\-levelwidesearch node\.wide Search Node \(depth=0\): expand along the brand dimension and delegate six brand\-level child nodes\.Web\-probing\-guided expansion: distributed by brandweb\-probing agent:evidence distribution =distributed; organization dimension =brand\. No single hub page covers all brands, so six brand\-level child nodes are delegated in parallel\.wide\(depth=1, Johnnie Walker\)parallel branch 1/6entity\_collect:collect the core/permanent product list for Johnnie Walker\.Output:Red Label, Black Label, …assign 8 atom agents:atom 1:collect attributes forRed Label\.atom 2:collect attributes forBlack Label\.experience:the official Johnnie Walker product pages consistently provide the needed fields, including product name…atom 3:collect attributes forDouble Black Label\.…wide\(depth=1, Grey Goose\)parallel branch 4/6entity\_collect:collect the core/permanent product list for Grey Goose\.Output:GREY GOOSE Vodka, Altius, …assign 3 atom agents:atom 1:collect attributes forGREY GOOSE Vodka\.atom 2:collect attributes forGREY GOOSE Altius\.…wide\(depth=1, Smirnoff\)parallel branch 3/6entity\_collect:collect the core/permanent product list for Smirnoff\.Output:Smirnoff White, No\.21 Vodka, …assign 3 atom agents:atom 1:collect attributes forSmirnoff White\.atom 2:collect attributes forNo\.21 Vodka\.…wide\(depth=1, Chivas Regal\)parallel branch 2/6entity\_collect:collect the core/permanent product list for Chivas Regal\.Output:Chivas Regal XV, Chivas Regal Extra, …assign 5 atom agents:atom 1:collect attributes forChivas Regal XV\.atom 2:collect attributes forChivas Regal Extra\.experience:search with the template “Chivas Regal \[Product Name\] product details specifications” to quickly locate information…atom 3:collect attributes forChivas Regal Ultis\.…wide\(depth=1, Absolut Vodka\)parallel branch 5/6entity\_collect:collect the core/permanent product list for Absolut Vodka\.Output:Absolut Elyx, Absolut Vodka, …assign 3 atom agents:atom 1:collect attributes forAbsolut Elyx\.atom 2:collect attributes forAbsolut Vodka\.…wide\(depth=1, Bacardi\)parallel branch 6/6entity\_collect:collect the core/permanent product list for Bacardi\.Output:Bacardi Gold, Bacardi Superior, …assign 7 atom agents:atom 1:collect attributes forBacardi Gold\.atom 2:collect attributes forBacardi Superior\.…Wide Node Result: return the merged Markdown table upward\.Root Node completionreceive the merged table and terminate the task\.BrandProductCategorySub\-categoryPack Size \(Bottle\)ABV %Johnnie WalkerRed LabelWhiskyBlended Scotch Whisky750ml40%………………

Figure 5:Case study for WS\_EN\_002 in the WideSearch benchmark\.Original QuestionCould you list every single concert on Taylor Swift’s official tour from January 1, 2010, to May 1, 2025, including the specific date, the concert’s English name, the country, the city, and the venue\. Each show should be on its own line, in chronological order from earliest to latest\.Please organize the results in one Markdown table with the following columns: Date, the Concert’s English Name, Host Country, Host City, Host Venue\. Do not use date ranges for Date; list it in the format of “Day Month, Year”, for example, 4th June, 2011\. Do not ask follow\-up questions; directly output the results according to the required columns\.

Recursive Wide Search DelegationRoot Nodereceive the original task, create a wide node, and assign it a task\.wide Node: List every single concert on Taylor Swift’s official tours from Jan\. 1, 2010 to May 1, 2025\. For each show, extract the specific date, concert English name, host country, host city, and venue\. Sort all rows chronologically and output one Markdown table without omitted cells\.web\-probing agent: web structure hint Coverage:comprehensive\.Sources:Wikipedia tour pages, Concert Archives, and tour\-level live\-performance summaries\.Data organization:sources are either tour\-based or reverse\-chronological\. The efficient strategy is to split by official tour, extract show\-level tables, filter dates, then merge chronologically\.Six tour\-level atom agents dispatched by wide and executed in parallelatom 1/6: Fearless Tourparallelsource:dedicated Wikipedia tour page,Fearless\_Tourtask:extract every official show within Jan\. 1, 2010–May 1, 2025\.atom 3/6: Speak Now World Tourparallelsource:dedicated Wikipedia tour page,Speak\_Now\_World\_Tourtask:extract every official show within Jan\. 1, 2010–May 1, 2025\.atom 5/6: The Red Tourparallelsource:dedicated Wikipedia tour page,The\_Red\_Tourtask:extract every official show within Jan\. 1, 2010–May 1, 2025\.atom 2/6: The 1989 World Tourparallelsource:dedicated Wikipedia tour page,The\_1989\_World\_Tourtask:extract every official show within Jan\. 1, 2010–May 1, 2025\.atom 4/6: Reputation Stadium Tourparallelsource:dedicated Wikipedia tour page,Reputation\_Stadium\_Tourtask:extract every official show within Jan\. 1, 2010–May 1, 2025\.atom 6/6: The Eras Tourparallelsource:dedicated Wikipedia tour page,The\_Eras\_Tourtask:extract every official show within Jan\. 1, 2010–May 1, 2025\.Wide Node Result: aggregate child\-node results into a table and return it upward\.Root Node completionreceive the returned chronological table and terminate the task\.DateConcert NameHost CountryHost CityHost Venue6th February, 2010Fearless TourAustraliaSydneyAcer Arena7th February, 2010Fearless TourAustraliaSydneyAcer Arena……………

Figure 6:Case study for WS\_EN\_006 in the WideSearch benchmark\.Original QuestionIdentify all US vehicles first launched or resumed production between 2010 and 2024, where the manufacturer is associated with a US entrepreneur who built experimental mobile devices, later pioneered continuous\-flow industrial production methods, and whose family\-owned business retains control through special voting rights\. For each standard launch edition, output launch MSRP, dimensions, wheelbase, torque, suspension, ADAS, and IPA in one Markdown table\.

Recursive Deep\-Wide Search DelegationRoot Nodefirst delegate hidden\-entity identification, and then collect vehicle table based on the returned evidence\.Deep Node: Identify the entrepreneur and associated manufacturer implied by the hidden clues: experimental vehicles, continuous\-flow industrial production, and family control through special voting rights\.searcher 1: Task: US entrepreneur vehicle manufacturing continuous\-flow industrial production methods / moving assembly line pioneer\.Output:✓Candidate identified: Henry Ford\.verifier: Task: verify Henry Ford’s experimental mobile devices before the Model T, including the 1896 Quadricycle, Sweepstakes racer, and pre\-Model\-T letter cars\.Output:✓Experimental\-vehicle clue matches Henry Ford\.verifier: Task: verify Ford Motor Company family control through special voting rights / Class B shares\.Output:✓Family\-control clue matches Ford\.Deep output:Henry Fordwide\(depth=0\): collect all eligible Ford models and dispatch one atom agent per vehicle to fill the requested columns\.entity\_collect:find all Ford vehicles first launched or resumed US production from 2010–2024, excluding facelifts/model updates\.Output:Focus Electric, C\-Max Energi, Fusion Energi, EcoSport, GT, Ranger, Bronco, Mustang Mach\-E, Maverick, F\-150 Lightning, Bronco Sport, Transit Connect, C\-Max Hybrid, F\-150 Raptor, Escape Plug\-in Hybrid, E\-Transitassign 16 atom agents:atom 1:collect attributes forFocus Electric \(2012\)\.atom 2:collect attributes forC\-Max Energi \(2013\)\.atom 3:collect attributes forFusion Energi \(2013\)\.…atom 4:collect attributes forEcoSport \(2018\)\.atom 5:collect attributes forGT \(2017\)\.atom 6:collect attributes forRanger \(2019\)\.Wide output: generate one Markdown table\.Root Node Completionreceive the returned Markdown table from the Wide node and terminate the task\.Model NamePriceDimensions L/W/HWheelbaseTorqueFront Susp\.Rear Susp\.ADASIPAFocus Electric \(2012\)$39,2004392/1823/14782649245MacPhersonMulti\-LinkRear Parking Sensor, ABS, …\-C\-Max Energi \(2013\)$32,9504409/1828/16212648239MacPhersonMulti\-LinkAdvanceTrac, traction control, …\-………………………

Figure 7:Case study for Wide2Deep\_ws\_en\_049 in the DeepWideSearch benchmark\.`Root Agent System Prompt`Figure 8:System prompt for guiding the root agent to create and dispatch subtasks to search nodes\.`Web\-Probing Agent Prompt`Figure 9:The system prompt for guiding the agent to probe Web information structures\.`Subtask Experience Extraction Prompt`Figure 10:The system prompt for guiding the LLM to summarize reusable subtask experience\.`Web Tool Interface`Figure 11:Descriptions of the Web\-Search Tool and the Page\-Browse Tool provided to agents\.