Recursive Multi-Agent Systems: Scaling Agent Collaboration Through Latent Recursion

Overview

Research area: NLP Authors: Xiyuan Yang, Jiaru Zou, Rui Pan, et al. Published: 2026-04-29 arXiv: 2504.21178

Summary

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. This work extends that scaling principle from a single model to multi-agent systems, asking: can agent collaboration itself be scaled through recursion?

To answer this, the authors introduce RecursiveMAS, a recursive multi-agent framework that casts the entire system as a unified latent-space recursive computation.

Key components

• RecursiveLink module: a lightweight module that connects heterogeneous agents into a collaboration loop, enabling in-distribution latent thought generation and cross-agent latent state transfer.
• Inner-outer loop learning algorithm: performs iterative whole-system co-optimization through shared gradient credit assignment across recursion rounds.

Theoretical results

• Runtime complexity analysis shows RecursiveMAS is more efficient than standard text-based multi-agent systems.
• Learning dynamics remain stable (consistent gradients) during recursive training.

Empirical results

Across 9 benchmarks in mathematics, science, medicine, search, and code generation, RecursiveMAS achieves:

• +8.3% average accuracy over state-of-the-art single-agent, multi-agent, and recursive-compute baselines
• 1.2x–2.4x end-to-end inference speedup
• 34.6%–75.6% reduction in token usage

Original Abstract

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...

---

*Auto-collected on 2026-04-30.*