Paper 2606.09730
SearchSwarm: Towards Delegation Intelligence in Agentic LLMs for Long-Horizon Deep Research
- Published
- Jun 2026
- Research lab
- Independent
- Citations
- 3
- GitHub
- 85 stars
01 In brief
Summary
SearchSwarm introduces a method for training delegation intelligence in LLM agents for long-horizon deep research tasks.
The approach uses a main-distributes, sub-executes paradigm where a main agent decomposes tasks and dispatches subtasks to subagents via a call_sub_agent tool, receiving only condensed, citation-grounded reports.
The harness design encourages delegation, comprehensive briefing, core judgment retention, and citation-grounded reporting.
Trajectories from this harness are used as supervised fine-tuning data to internalize delegation behavior into model weights.
The resulting model, SearchSwarm-30B-A3B, achieves state-of-the-art results among comparable-scale models: 68.1 on BrowseComp, 73.3 on BrowseComp-ZH, 82.5 on GAIA, and 80.8 on xbench-DeepSearch-2505, outperforming larger models like GPT-5.2-Thinking and DeepSeek V3.2 on some benchmarks.
Ablations show the harness improves performance by 10 points over the base framework, and the training data generalizes to single-agent settings and open-ended research tasks.
The authors release the harness, model weights, and training data to facilitate future research.
02 From the paper
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.