WideSeek-R1: Exploring Width Scaling for Broad Information Seeking via Multi-Agent Reinforcement Learning

Papers with Code Trending Papers

Summary

WideSeek-R1 is a multi-agent reinforcement learning framework designed for broad information seeking, exploring width scaling to achieve performance comparable to larger single agents with fewer parameters.

Recent advancements in Large Language Models (LLMs) have largely focused on depth scaling, where a single agent solves long-horizon problems with multi-turn reasoning and tool use. However, as tasks grow broader, the key bottleneck shifts from individual competence to organizational capability. In this work, we explore a complementary dimension of width scaling with multi-agent systems to address broad information seeking. Existing multi-agent systems often rely on hand-crafted workflows and turn-taking interactions that fail to parallelize work effectively. To bridge this gap, we propose WideSeek-R1, a lead-agent-subagent framework trained via multi-agent reinforcement learning (MARL) to synergize scalable orchestration and parallel execution. By utilizing a shared LLM with isolated contexts and specialized tools, WideSeek-R1 jointly optimizes the lead agent and parallel subagents on a curated dataset of 20k broad information-seeking tasks. Extensive experiments show that WideSeek-R1-4B achieves an item F1 score of 40.0% on the WideSearch benchmark, which is comparable to the performance of single-agent DeepSeek-R1-671B. Furthermore, WideSeek-R1-4B exhibits consistent performance gains as the number of parallel subagents increases, highlighting the effectiveness of width scaling.
Original Article
View Cached Full Text

Cached at: 09/07/26, 06:06 PM

Paper page - WideSeek-R1: Exploring Width Scaling for Broad Information Seeking via Multi-Agent Reinforcement Learning

Source: https://huggingface.co/papers/2602.04634

Abstract

Multi-agent systems using reinforcement learning enable parallel information seeking with scalable orchestration, achieving performance comparable to larger single agents.

Recent advancements inLarge Language Models(LLMs) have largely focused on depth scaling, where a single agent solves long-horizon problems with multi-turn reasoning and tool use. However, as tasks grow broader, the key bottleneck shifts from individual competence to organizational capability. In this work, we explore a complementary dimension of width scaling withmulti-agent systemsto address broadinformation seeking. Existingmulti-agent systemsoften rely on hand-crafted workflows and turn-taking interactions that fail to parallelize work effectively. To bridge this gap, we propose WideSeek-R1, alead-agent-subagent frameworktrained viamulti-agent reinforcement learning(MARL) to synergize scalable orchestration andparallel execution. By utilizing a shared LLM with isolated contexts and specialized tools, WideSeek-R1 jointly optimizes the lead agent and parallel subagents on a curated dataset of 20k broad information-seeking tasks. Extensive experiments show that WideSeek-R1-4B achieves an itemF1 scoreof 40.0% on theWideSearch benchmark, which is comparable to the performance of single-agent DeepSeek-R1-671B. Furthermore, WideSeek-R1-4B exhibits consistent performance gains as the number of parallel subagents increases, highlighting the effectiveness of width scaling.

View arXiv pageView PDFProject pageGitHub4.84kAdd to collection

Get this paper in your agent:

hf papers read 2602\.04634

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

Models citing this paper1

#### RLinf/WideSeek-R1-4b Text Generation• 4B• UpdatedMar 13 • 20 • 6

Datasets citing this paper3

#### RLinf/WideSeek-R1-train-data Preview• UpdatedMar 13 • 627 • 2 #### RLinf/Wiki-2018-Corpus #### RLinf/WideSeek-R1-test-data Viewer• UpdatedMar 13 • 200 • 47

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2602.04634 in a Space README.md to link it from this page.

Collections including this paper3

Similar Articles

OpenWebRL: Demystifying Online Multi-turn Reinforcement Learning for Visual Web Agents

Hugging Face Daily Papers

OpenWebRL presents an open framework for training visual web agents using online multi-turn reinforcement learning on real websites, achieving state-of-the-art performance with minimal initial supervision. Their 4B-parameter model outperforms prior open agents and competes with proprietary systems like OpenAI CUA and Gemini CUA.