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

2026-02-04

AI Summary

This work addresses the limited collaboration and parallelism inherent in single-agent deep reasoning for wide-area information retrieval by proposing WideSeek, a master-slave multi-agent framework. WideSeek introduces the concept of breadth expansion into large language model systems, leveraging multi-agent reinforcement learning (MARL) to jointly optimize the master agent’s task decomposition and the parallel execution of multiple sub-agents. The framework integrates shared models, isolated contexts, and specialized tools, overcoming the constraints of traditional handcrafted workflows. Evaluated on the WideSearch benchmark, WideSeek-R1-4B achieves an entry-level F1 score of 40.0%, matching the performance of the much larger single-agent DeepSeek-R1-671B, with consistent performance gains as the number of sub-agents increases.

Technology Category

• Multiagent Systems: Multiagent Learning
• Search and Optimization: Learning to Search
• Machine Learning: Large Multimodal Models (LMMs)

Application Category

• Search and Retrieval-Augmented AI: Agentic search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

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.

Problem

• width scaling
• broad information seeking
• multi-agent systems
• parallel execution
• organizational capability

Innovation

• width scaling
• multi-agent reinforcement learning
• parallel execution
• broad information seeking
• lead-agent-subagent framework