Agent Q-Mix: Selecting the Right Action for LLM Multi-Agent Systems through Reinforcement Learning

TL;DR

Agent Q-Mix introduces a reinforcement learning framework that optimizes multi-agent communication topologies for large language model systems, improving accuracy and efficiency across various reasoning tasks.

Contribution

It reformulates topology selection as a cooperative MARL problem, integrating GNNs and value factorization for decentralized communication decisions in LLM multi-agent systems.

Findings

Achieves highest average accuracy on seven benchmarks.

Demonstrates superior token efficiency and robustness.

Outperforms existing frameworks on Humanity's Last Exam.

Abstract

Large Language Models (LLMs) have shown remarkable performance in completing various tasks. However, solving complex problems often requires the coordination of multiple agents, raising a fundamental question: how to effectively select and interconnect these agents. In this paper, we propose \textbf{Agent Q-Mix}, a reinforcement learning framework that reformulates topology selection as a cooperative Multi-Agent Reinforcement Learning (MARL) problem. Our method learns decentralized communication decisions using QMIX value factorization, where each agent selects from a set of communication actions that jointly induce a round-wise communication graph. At its core, Agent Q-Mix combines a topology-aware GNN encoder, GRU memory, and per-agent Q-heads under a Centralized Training with Decentralized Execution (CTDE) paradigm. The framework optimizes a reward function that balances task accuracy with token cost. Across seven core benchmarks in coding, reasoning, and mathematics, Agent Q-Mix achieves the highest average accuracy compared to existing methods while demonstrating superior token efficiency and robustness against agent failure. Notably, on the challenging Humanity's Last Exam (HLE) using Gemini-3.1-Flash-Lite as a backbone, Agent Q-Mix achieves 20.8\% accuracy, outperforming Microsoft Agent Framework (19.2\%) and LangGraph (19.2\%), followed by AutoGen and Lobster by OpenClaw. These results underscore the effectiveness of learned, decentralized topology optimization in pushing the boundaries of multi-agent reasoning.