Behavior Knowledge Merge in Reinforced Agentic Models

TL;DR

This paper introduces Reinforced Agent Merging (RAM), a novel method for effectively merging RL-trained agentic models by addressing task-vector mismatch issues, leading to improved performance over existing methods.

Contribution

The paper proposes RAM, a distribution-aware merging framework that preserves task-specific behaviors in RL-trained models, outperforming traditional merging approaches.

Findings

RAM outperforms baseline merging methods across multiple domains.

RAM enables synergistic effects, surpassing individual specialized agents.

The method effectively preserves task-specific capabilities during merging.

Abstract

Reinforcement learning (RL) is central to post-training, particularly for agentic models that require specialized reasoning behaviors. In this setting, model merging offers a practical mechanism for integrating multiple RL-trained agents from different tasks into a single generalist model. However, existing merging methods are designed for supervised fine-tuning (SFT), and they are suboptimal to preserve task-specific capabilities on RL-trained agentic models. The root is a task-vector mismatch between RL and SFT: on-policy RL induces task vectors that are highly sparse and heterogeneous, whereas SFT-style merging implicitly assumes dense and globally comparable task vectors. When standard global averaging is applied under this mismatch, RL's non-overlapping task vectors that encode critical task-specific behaviors are reduced and parameter updates are diluted. To address this issue, we propose Reinforced Agent Merging (RAM), a distribution-aware merging framework explicitly designed for RL-trained agentic models. RAM disentangles shared and task-specific unique parameter updates, averaging shared components while selectively preserving and rescaling unique ones to counteract parameter update dilution. Experiments across multiple agent domains and model architectures demonstrate that RAM not only surpasses merging baselines, but also unlocks synergistic potential among agents to achieve performance superior to that of specialized agents in their domains.