Towards Foundation Models with Native Multi-Agent Intelligence

2025-12-09

AI Summary

Current foundation models (FMs) exhibit strong single-agent capabilities but lack native multi-agent collaboration—systematically deficient in joint perception, cooperative planning, efficient communication, and dynamic adaptation. Method: We conduct the first large-scale empirical analysis across 41 large language models, demonstrating that single-agent performance does not transfer to multi-agent settings. Building on this finding, we propose the first unified framework for natively multi-agent intelligence, comprising a dedicated training paradigm, multi-role interactive data construction, a task-decoupled evaluation suite, and collaborative safety mechanisms. Contribution/Results: Experiments show that our framework significantly improves robustness and generalization on multi-agent tasks. It exposes fundamental limitations of existing FMs and establishes a reproducible technical pathway and empirical benchmark for next-generation multi-agent foundation models.

Technology Category

• Multiagent Systems: Other Foundations of Multi Agent Systems
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Safety and Robustness

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Foundation models (FMs) are increasingly assuming the role of the "brain" of AI agents. While recent efforts have begun to equip FMs with native single-agent abilities -- such as GUI interaction or integrated tool use -- we argue that the next frontier is endowing FMs with native multi-agent intelligence. We identify four core capabilities of FMs in multi-agent contexts: understanding, planning, efficient communication, and adaptation. Contrary to assumptions about the spontaneous emergence of such abilities, we provide extensive empirical evidence across 41 large language models showing that strong single-agent performance alone does not automatically yield robust multi-agent intelligence. To address this gap, we outline key research directions -- spanning dataset construction, evaluation, training paradigms, and safety considerations -- for building FMs with native multi-agent intelligence.

Problem

• Develops foundation models with native multi-agent intelligence capabilities.
• Addresses the gap between single-agent and multi-agent performance in AI.
• Proposes research directions for multi-agent dataset, training, and safety.

Innovation

• Endowing foundation models with native multi-agent intelligence
• Identifying core multi-agent capabilities like planning and adaptation
• Proposing research directions for dataset construction and training