SearchSwarm: Towards Delegation Intelligence in Agentic LLMs for Long-Horizon Deep Research

2026-06-08

AI Summary

This work addresses the challenge that large language models, constrained by limited context windows, struggle to effectively decompose complex long-horizon tasks and integrate their results. The authors propose a guided agent framework that, for the first time in the open-source community, systematically synthesizes high-quality task decomposition and delegation trajectories as supervised fine-tuning data, thereby internalizing delegation capabilities directly into model parameters. By integrating multi-agent collaboration and context compression strategies, the framework enables the model to autonomously decide when to delegate subtasks and how to synthesize their outcomes. The resulting SearchSwarm-30B-A3B model achieves state-of-the-art performance among models of comparable scale, scoring 68.1 on BrowseComp and 73.3 on BrowseComp-ZH.

Abstract

Large language models are increasingly expected to handle complex, long-horizon real-world tasks whose context demands can grow without bound, yet model context windows remain inherently finite. Recent work explores a paradigm where a main agent decomposes tasks and dispatches subtasks to subagents, which execute and return only summarized results, conserving the main agent's context budget. However, performing this well requires delegation intelligence: the ability to decompose complex tasks, determine when and what to delegate, and integrate returned results into the ongoing workflow. Training data for this capability is scarce in naturally occurring text, and to our knowledge, how to synthesize such data and train models to acquire this capability remains largely unexplored in the open-source community. To bridge this gap, we present a preliminary exploration targeting deep research, a representative long-horizon agent task. Specifically, we design a harness that guides the model toward high-quality task decomposition and delegation, while constraining subagents to return results properly to support the main agent's workflow. The harness-guided trajectories naturally encode correct delegation decisions, which we use as supervised fine-tuning data to internalize delegation intelligence into model weights. Our resulting model, SearchSwarm-30B-A3B, achieves 68.1 on BrowseComp and 73.3 on BrowseComp-ZH, the best results among all models of comparable scale. We will release our harness, model weights, and training data to facilitate future research.

Problem

• delegation intelligence
• long-horizon tasks
• agentic LLMs
• context window limitation
• deep research

Innovation

• delegation intelligence
• agentic LLMs
• task decomposition
• context budgeting
• supervised fine-tuning