Multi-Agent Deep Research: Training Multi-Agent Systems with M-GRPO

TL;DR

This paper introduces M-GRPO, a hierarchical training method for multi-agent systems with distinct LLMs, improving stability and efficiency in tool-augmented reasoning tasks through trajectory alignment and decoupled optimization.

Contribution

The paper proposes M-GRPO, a hierarchical policy optimization algorithm that enables scalable training of multi-agent systems with separate LLMs, addressing optimization challenges in distributed environments.

Findings

M-GRPO outperforms single-agent and frozen sub-agent baselines on real-world benchmarks.

Decoupled training pipeline enables scalable multi-agent training across servers.

Trajectory alignment improves stability and sample efficiency in multi-agent reasoning.

Abstract

Multi-agent systems perform well on general reasoning tasks. However, the lack of training in specialized areas hinders their accuracy. Current training methods train a unified large language model (LLM) for all agents in the system. This may limit the performances due to different distributions underlying for different agents. Therefore, training multi-agent systems with distinct LLMs should be the next step to solve. However, this approach introduces optimization challenges. For example, agents operate at different frequencies, rollouts involve varying sub-agent invocations, and agents are often deployed across separate servers, disrupting end-to-end gradient flow. To address these issues, we propose M-GRPO, a hierarchical extension of Group Relative Policy Optimization designed for vertical Multi-agent systems with a main agent (planner) and multiple sub-agents (multi-turn tool executors). M-GRPO computes group-relative advantages for both main and sub-agents, maintaining hierarchical credit assignment. It also introduces a trajectory-alignment scheme that generates fixed-size batches despite variable sub-agent invocations. We deploy a decoupled training pipeline in which agents run on separate servers and exchange minimal statistics via a shared store. This enables scalable training without cross-server backpropagation. In experiments on real-world benchmarks (e.g., GAIA, XBench-DeepSearch, and WebWalkerQA), M-GRPO consistently outperforms both single-agent GRPO and multi-agent GRPO with frozen sub-agents, demonstrating improved stability and sample efficiency. These results show that aligning heterogeneous trajectories and decoupling optimization across specialized agents enhances tool-augmented reasoning tasks.