Recursive Multi-Agent Systems

Papers with Code Trending Papers
multi-agent-systems recursive-computation latent-space efficiency collaborative-reasoning huggingface

Summary

This paper introduces RecursiveMAS, a framework that extends recursive scaling principles to multi-agent systems for improved collaborative reasoning efficiency and accuracy. It demonstrates significant speedups and token reduction across various benchmarks compared to standard baselines.

Recursive or looped language models have recently emerged as a new scaling axis by iteratively refining the same model computation over latent states to deepen reasoning. We extend such scaling principle from a single model to multi-agent systems, and ask: Can agent collaboration itself be scaled through recursion? To this end, we introduce RecursiveMAS, a recursive multi-agent framework that casts the entire system as a unified latent-space recursive computation. RecursiveMAS connects heterogeneous agents as a collaboration loop through the lightweight RecursiveLink module, enabling in-distribution latent thoughts generation and cross-agent latent state transfer. To optimize our framework, we develop an inner-outer loop learning algorithm for iterative whole-system co-optimization through shared gradient-based credit assignment across recursion rounds. Theoretical analyses of runtime complexity and learning dynamics establish that RecursiveMAS is more efficient than standard text-based MAS and maintains stable gradients during recursive training. Empirically, we instantiate RecursiveMAS under 4 representative agent collaboration patterns and evaluate across 9 benchmarks spanning mathematics, science, medicine, search, and code generation. In comparison with advanced single/multi-agent and recursive computation baselines, RecursiveMAS consistently delivers an average accuracy improvement of 8.3%, together with 1.2times-2.4times end-to-end inference speedup, and 34.6%-75.6% token usage reduction. Code and Data are provided in https://recursivemas.github.io.
Original Article
View Cached Full Text

Cached at: 05/08/26, 08:59 AM

Paper page - Recursive Multi-Agent Systems

Source: https://huggingface.co/papers/2604.25917

Abstract

RecursiveMAS extends recursive scaling principles from single models to multi-agent systems, enabling collaborative reasoning through iterative latent-space computations with improved efficiency and accuracy.

Recursive or looped language models have recently emerged as a new scaling axis by iteratively refining the same model computation over latent states to deepen reasoning. We extend such scaling principle from a single model tomulti-agent systems, and ask: Can agent collaboration itself be scaled through recursion? To this end, we introduce RecursiveMAS, a recursive multi-agent framework that casts the entire system as a unifiedlatent-space recursive computation. RecursiveMAS connects heterogeneous agents as a collaboration loop through the lightweightRecursiveLink module, enabling in-distribution latent thoughts generation and cross-agent latent state transfer. To optimize our framework, we develop aninner-outer loop learningalgorithm for iterative whole-system co-optimization through sharedgradient-based credit assignmentacross recursion rounds. Theoretical analyses ofruntime complexityandlearning dynamicsestablish that RecursiveMAS is more efficient than standard text-based MAS and maintains stable gradients during recursive training. Empirically, we instantiate RecursiveMAS under 4 representativeagent collaboration patternsand evaluate across 9 benchmarks spanning mathematics, science, medicine, search, and code generation. In comparison with advanced single/multi-agent and recursive computation baselines, RecursiveMAS consistently delivers an average accuracy improvement of 8.3%, together with 1.2times-2.4timesend-to-end inference speedup, and 34.6%-75.6%token usage reduction. Code and Data are provided in https://recursivemas.github.io.

View arXiv pageView PDFProject pageGitHub308Add to collection

Get this paper in your agent:

hf papers read 2604\.25917

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper19

#### RecursiveMAS/Mixture-Science-BioMistral-7B 7B• Updated9 days ago • 60 • 2 #### RecursiveMAS/Mixture-Summarizer-Qwen3.5-2B 2B• Updated9 days ago • 90 • 2 #### RecursiveMAS/Deliberation-Reflector-Qwen3.5-4B 4B• Updated9 days ago • 50 • 2 #### RecursiveMAS/Deliberation-Toolcaller-Qwen3.5-4B 4B• Updated9 days ago • 47 • 1 Browse 19 models citing this paper## Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2604.25917 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2604.25917 in a Space README.md to link it from this page.

Collections including this paper15

Browse 15 collections that include this paper

Similar Articles

TMAS: Scaling Test-Time Compute via Multi-Agent Synergy

Hugging Face Daily Papers

TMAS introduces a multi-agent framework that enhances large language model reasoning by scaling test-time compute through structured collaboration and hierarchical memory systems. The approach uses specialized agents, cross-trajectory information flow, and hybrid reward reinforcement learning to improve iterative scaling and stability on challenging reasoning benchmarks.

DuMate-DeepResearch: An Auditable Multi-Agent System with Recursive Search and Rubric-Grounded Reasoning

arXiv cs.AI

This technical report introduces DuMate-DeepResearch, a multi-agent framework for deep research tasks that decouples the agent core from a tool ecosystem, and incorporates graph-based dynamic planning, recursive two-level execution, and rubric-based test-time optimization. The system achieves state-of-the-art results on two deep research benchmarks, demonstrating the value of auditable agent infrastructure.

NeuroMAS: Multi-Agent Systems as Neural Networks with Joint Reinforcement Learning

arXiv cs.AI

NeuroMAS treats multi-agent language systems as trainable neural-network-like architectures with LLM agents as nodes, using reinforcement learning to learn communication and specialization. It shows improved performance and that progressive growth from smaller systems works better than training large systems from scratch.