SearchSwarm: Towards Delegation Intelligence in Agentic LLMs for Long-Horizon Deep Research
English summary
Researchers introduce SearchSwarm-30B-A3B, an agentic LLM designed for long-horizon research tasks. It employs delegation intelligence to decompose complex problems, delegate subtasks to subagents, and integrate summarized results, thereby optimizing the main agent's context budget. Because natural training data is scarce, the team synthesized data and used a harness to guide task decomposition and subagent coordination. The model outperforms similarly sized counterparts and will be open-sourced to foster further investigation.
Chinese summary
研究人员提出SearchSwarm-30B-A3B，一个面向长周期研究任务的代理型大语言模型。该模型运用委托智能将复杂问题分解、委派给子代理并整合摘要结果，从而优化主代理的语境预算。由于自然训练数据稀缺，团队通过数据合成与引导式框架实现任务分解和子代理协调。该模型表现优于同规模模型，并将开源以推动后续研究。
Key points
SearchSwarm-30B-A3B uses delegation intelligence to break down long-horizon tasks, coordinate subagents, and merge results.
SearchSwarm-30B-A3B 通过委托智能分解长周期任务、协调子代理并整合结果。
Training leveraged synthesized data and a harness framework to overcome the lack of natural delegation data.
训练中采用合成数据与引导式框架，解决了自然委托数据不足的问题。
The model achieves superior performance compared to similar-scale models and will be released open-source.
该模型性能优于同规模模型，并将开源共享。