Source record

Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2610.08155.

Pipeline notes

The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.

Paper proposes S1-MAS, a token-efficient multi-agent framework that assigns bounded coordination decisions to lightweight System One models while reserving open-ended reasoning for capable LLM workers, reducing GPT-4o token consumption by 44.9%-97.2% and latency by 37.8%-93.0% across seven benchmarks versus AgentVerse, DyLAN, and SelfOrg.

Decoupling cheap coordination from expensive reasoning via a lightweight controller is a concrete architectural pattern with measured cost savings across three baselines — directly applicable to anyone scaling multi-agent workflows.

Cited by

No citations on record.