AgentArk: Distilling Multi-Agent Intelligence into a Single LLM Agent

Feb 3, 2026·
Yinyi Luo
Yiqiao Jin
Yiqiao Jin
,
Weichen Yu
,
Mengqi Zhang
,
Srijan Kumar
,
Xiaoxiao Li
,
Weijie Xu
,
Xin Chen
,
Jindong Wang
· 1 min read
Abstract
AgentArk transfers the reasoning dynamics of multi-agent debate into a single language model. It studies reasoning-enhanced fine-tuning, trajectory-based augmentation, and process-aware distillation, moving computation from inference to training. The resulting agent aims to retain the reasoning, self-correction, and generalization benefits of multi-agent interaction with the inference cost of one model.
Type
Publication
NeurIPS 2026

Overview

AgentArk transfers multi-agent reasoning into a single language model through reasoning-enhanced fine-tuning, trajectory-based augmentation, and process-aware distillation. It studies how moving computation into training affects reasoning, self-correction, robustness, and generalization.

LLM Agents Multi-Agent Systems Knowledge Distillation Efficient AI
Yiqiao Jin
Authors
Yiqiao Jin
Ph.D. Candidate in Computer Science
My research sits at the intersection of multimodal foundation models, intelligent agents, and spatial/physical intelligence. I build general-purpose models that perceive, reason, learn, and act in interactive virtual and physical environments.