# Streaming Communication in Multi-Agent Reasoning

DevFeed: [Streaming Communication in Multi-Agent Reasoning](<https://devfeed.tech/articles/streaming-communication-in-multi-agent-reasoning-102974.md>)

Original publisher: [Read original article](<https://arxiv.org/abs/2606.05158>)

Author: Zhen Yang, Xiaogang Xu, Wen Wang, Cong Chen, Xander Xu, Ying-Cong Chen

Published: 2026-10-28T04:00:00Z

Content type: article

Language: en

Sources: [arXiv: Artificial Intelligence](<https://devfeed.tech/sources/arxiv-artificial-intelligence.md>)

Topics: [Multi Agent Systems](<https://devfeed.tech/topics/multi-agent-systems.md>), [agent to agent communication](<https://devfeed.tech/topics/agent-to-agent-communication.md>), [llm-reasoning](<https://devfeed.tech/topics/llm-reasoning.md>)

Tags: [claude-opus](<https://devfeed.tech/tags/claude-opus.md>), [code](<https://devfeed.tech/tags/code.md>), [cs-ai](<https://devfeed.tech/tags/cs-ai.md>), [cs-cl](<https://devfeed.tech/tags/cs-cl.md>), [cs-ma](<https://devfeed.tech/tags/cs-ma.md>), [efficiency](<https://devfeed.tech/tags/efficiency.md>), [glm-5-2](<https://devfeed.tech/tags/glm-5-2.md>), [gpt-5-4](<https://devfeed.tech/tags/gpt-5-4.md>), [graph](<https://devfeed.tech/tags/graph.md>), [latency](<https://devfeed.tech/tags/latency.md>), [llms](<https://devfeed.tech/tags/llms.md>), [mathematics](<https://devfeed.tech/tags/mathematics.md>), [multi-agent](<https://devfeed.tech/tags/multi-agent.md>), [pipelining](<https://devfeed.tech/tags/pipelining.md>), [reasoning](<https://devfeed.tech/tags/reasoning.md>), [scaling](<https://devfeed.tech/tags/scaling.md>), [science](<https://devfeed.tech/tags/science.md>), [streaming](<https://devfeed.tech/tags/streaming.md>)

## AI overview

StreamMA reduces multi-agent reasoning latency by sending each generated reasoning step downstream immediately, allowing adjacent agents to work in parallel. The authors say early reasoning steps are more reliable than later ones, so streaming them can also improve effectiveness by limiting downstream exposure to error-prone late steps. Their closed-form analysis compares streaming, serial, and single-agent protocols. Across eight math, science, and code benchmarks, three LLMs, and Chain, Tree, and Graph topologies, StreamMA outperformed both baselines by an average of 7.3 percentage points, with a reported maximum gain of 22.4 points on HMMT2026 using Claude Opus 4.6-High. The abstract also reports that increasing steps per agent improved both effectiveness and efficiency; it provides no further detail on the bounds, cost ratios, or experimental setup.

## Source excerpt

arXiv:2606.05158v3 Announce Type: replace-cross Abstract: Multi-agent reasoning systems adopt a generate-then-transfer paradigm that forces end-to-end latency to scale linearly with pipeline depth. We introduce StreamMA, a multi-agent reasoning system that streams each reasoning step to downstream agents as soon as it is generated, pipelining adjacent agents and thus reducing latency. Surprisingly, this pipelining also improves effectiveness: because multi-step reasoning quality is non-uniform and early steps are more reliable than later ones, working with these reliable early steps instead of the full chain prevents error-prone late steps from misleading downstream agents. We formalize both advantages with the first closed-form joint analysis of stream, serial, and single protocols, deriving the effectiveness ordering, speedup upper bound, and cost ratio. Across eight reasoning benchmarks spanning mathematics, science, and code, three frontier LLMs (Claude Opus 4.6, GPT-5.4, and GLM-5.2), and three topologies (Chain, Tree, Graph), StreamMA outperforms both baselines (avg.+7.3pp, max +22.4pp on HMMT2026; Claude Opus 4.6-High). Beyond these contributions, we discover a step-level scaling law: increasing per-agent steps consistently improves both effectiveness and efficiency, a new scaling dimension orthogonal to and composable with agent-count scaling.