Neuro-Inspired Inhibitory Message Passing for Accurate and Token-Efficient Multi-Agent Reasoning

Nima H. Siboni; Magdalena Springer; Vahid Rostami

cs.AIOct 17, 2026

Neuro-Inspired Inhibitory Message Passing for Accurate and Token-Efficient Multi-Agent Reasoning

Authors: Nima H. Siboni, Magdalena Springer, Vahid Rostami

Organizations: Computational Systems Neuroscience, Institute of Zoology, University of Cologne, Germany

Abstract

Multi-agent systems promise to solve problems by bringing together information beyond the reach of any individual agent. Retrieval and tool use can distribute this information across agents, making communication essential to collaboration. A central challenge is keeping crucial evidence visible in each agent's context as messages accumulate. Repeated information can crowd out rare, decisive evidence. Models of cortical circuits suggest a principle for addressing this imbalance: inhibition can selectively suppress shared inputs while preserving sensitivity to weaker signals. We adapt this principle into inhibitory message passing, a training-free communication router that cancels redundant evidence separately for each evidence sentence and receiver. Each round, it forwards evidence sentences restated in agents' messages, delivering each at most once to receivers whose own evidence lacks it and excluding the senders' reasoning and answers. Against standard message passing, accuracy in peer groups rose by 39-77 percentage points across three benchmarks, and tokens per run fell by 10-46%. Gains held across seven open models spanning four to thirty-two billion parameters and two commercial models, in groups up to sixty-four agents, and across multi-agent architectures including peer-to-peer, orchestrator-worker, and multilevel hierarchical organizations. These findings support a scalable design pattern for more accurate and token-efficient multi-agent reasoning.

Figures & tables

Figure 1: Inhibitory message passing gives each agent the distributed evidence it lacks, once per round, and cancels what it already has. ( a ) An illustrative task with distributed evidence. Every agent reads the chart (shared evidence), and three agents’ own tools each return one result (private evidence, one holder each). The shared evidence suggests a heart attack; only the private evidence, combined, points to an aortic dissection. ( b ) The same messages, delivered two ways to a receiver, which can be a peer, a lead or a head. Under standard message passing the receiver gets every message as written, with its answer. Under inhibitory message passing a router forwards each restated private evidence sentence once per round to every receiver that lacks it ( Equation 1 ); the nurse’s result has not been restated yet. ( c ) Inhibition tracks excitation input by input, in a circuit model (top) and in the router (bottom; Equation 2 ): what the receiver already has is cancelled (

\theta=0

), and each evidence sentence it lacks passes once (

\theta=1

). As inputs are added, excitation

E

and inhibition

I

grow together while

E-I

stays flat. ( d ) The expected consequence, sketched for agents that hear every sender and lack a fixed set of evidence: as the group grows, standard message passing drifts toward more input and lower accuracy, while inhibitory message passing holds each agent at one operating point: the same, smaller input and the same, higher accuracy at every

N

. Figure 3 a tests the input, Figure 2 the accuracy. Orange: standard message passing; blue: inhibitory message passing; gray: shared evidence; red: private evidence.

Extension	What changes	Question it answers	Shown in
Core comparison	Standard vs. inhibitory message passing in peer groups; Qwen3-8B, Olmo-3-7B; HiddenBench, decoy puzzles, MuSiQue; $N=4$ –32	Does the router help a peer group?	Fig. 2 a–c
Group size	$N$ up to 64 on HiddenBench, same evidence	Does accuracy hold as the group grows?	Figs. 2 a, 3 a
Model	Ministral-3-8B ( Liu et al., 2026 ) and Gemma-4-12B ( Gemma Team et al., 2026 ) (HiddenBench, puzzles); Qwen3 at 4B, 14B and 32B, GPT-6 Luna, Claude Haiku 5.5 (HiddenBench)	Is the gain one model’s weakness?	Fig. 2 d; App. E
Architecture	Orchestrator–worker and hierarchical teams (a lead hears its staff), beside peer groups that hear every peer; HiddenBench, puzzles; $N=8$ –32	Does the router matter when one agent decides?	Fig. 2 e
Check: own searches	Holders set by each agent’s own search; MuSiQue; $N=32$ ; peer groups and a lead that delegates nothing	Did assigning the holders do the work?	App. G
Taking the router apart	One rule of Eq. ( 1 ) removed, or the threshold of Eq. ( 2 ) moved or replaced by a random sample of the same size; a full inbox history; an LLM summary of messages; $N=4$ –32	Which of the three rules carries the gain?	Figs. 3 b, 4 , Table H.3 ; App. H

Table 1: One core comparison, extended one axis at a time. Each row keeps the core comparison’s settings except those listed; the model and architecture extensions and the search check also change the benchmarks and the group sizes. Every model ran with thinking off ( Sections A.2 and E ).

N

: agents per group (the workers, in teams).

Figure 2: Inhibitory message passing raised group accuracy across group size, model and architecture, at fewer tokens. ( a – c ) Group accuracy against

N

, two main models; the prior-work baseline ran up to

N=32

. ( d ) HiddenBench at

N=16

: seven open models; below the rule, two commercial models. ( e ) Three architectures, group accuracy averaged over

N=8

–32. Each line runs from standard (orange) to inhibitory message passing (blue). Circles: Qwen3-8B; squares: Olmo-3-7B. Dark boxes mark where the group’s answer is decided: a vote, the lead or the head. Here peer-to-peer groups hear every peer, unlike a–c. ( f ) Tokens saved per run by inhibitory message passing, in % of standard message passing’s tokens, counting every agent, the decider included. Dots: settings, for peer-to-peer those of a–c; bars: means. Bars in a–e: 95% intervals.

Figure 3: Every holder within reach without the repeated input: cancellation absorbs the copies that hearing every sender brings, and six senders lose reach as the group grows. ( a ) HiddenBench, inhibitory message passing: tokens of the evidence sentences per inbox that arrive from every sender (grey) and that are delivered (blue; Equation 2 , summed over sentences); the difference is cancelled because the receiver already holds those sentences, the scaling sketched in Figure 1 c. ( b ) Group accuracy of inhibitory message passing hearing every sender (blue, as in Figure 2 a–b) or six random senders (green, 95% intervals); below eight agents, six senders are every sender. ( c ) Reach of six senders relative to every sender on the three benchmarks, against

\rho(N)

of Equation 3 : six senders lose exactly the reach that a random draw of six predicts. Solid lines and circles: Qwen3-8B; dashed lines and squares: Olmo-3-7B.

Figure 4: Each of the router’s three rules is needed. ( a ) One round. Agents write and read messages as under standard message passing; the router delivers the original evidence sentences their messages restate (1), one copy of each (2), only to receivers that lack them (3), and nothing of the messages themselves. ( b ) Variants that drop one or more rules (open circles) lose group accuracy against inhibitory message passing: two main models at

N=16

and 32, paired on items; bars, mean; lines, range. Deduplication only, the random sample and the LLM summary are cut to the router’s inbox size.

Appendix figures & tables21 assets

Supplementary material from the paper’s appendix.

Appendix

Benchmark	Shared evidence	Private evidence	Items (development / test)	Answer
HiddenBench	three to eight sentences (median four) that favour one wrong option, the decoy	three or four sentences	24 / 41 tasks	one of three or four options
Decoy puzzles	roads that lead to a wrong landmark (strong decoy) or to none (weak decoy)	four roads	48 / 100 items per decoy strength	the landmark reached
MuSiQue-Ans	four paragraphs that do not answer the question	two to four supporting paragraphs	screened per model	free text, scored by normalized exact match

Appendix

Table A.1: Three benchmarks, one construction. Every agent holds the shared evidence; each private evidence sentence has one holder.

Condition	Who a receiver hears	What a receiver sees
no communication	nobody	its own evidence
full information	nobody	all evidence and its last message; “one answer”: all evidence, answering afresh each round
standard message passing	6 random peers	their latest messages, as written
inhibitory message passing	every peer with a parsed answer	evidence sentences that peers’ messages restate and it lacks, each once per round; no answers
exchange-then-decide	6 random peers	messages; votes only after the last round
one message per answer	one peer per distinct answer	messages as written

Appendix

Table A.2: Conditions. The last four rows adapt published remedies: exchange-then-decide and share-all from HiddenBench ( Li et al., 2026b ) (the share-all wording is ours), one message per answer from the answer-level filter that diversity-aware message retention (DAR) tested beside its own ( Nguyen et al., 2026 ) , and anonymous message passing from an anonymous evidence board ( Ji et al., 2026 ) . “6 senders” in a table marks the router’s inbox built from six random peers.

Round 0: each agent answers alone

\langle

task description

\rangle

Information available to you: -

\langle

evidence sentence

\rangle

\langle

evidence sentence

\rangle

… Options:

\langle

option

\rangle

\langle

option

\rangle

, … Which option do you choose? Reason in at most 100 words, then end with a last line of the form "Final answer:

\langle

option

\rangle

Rounds 1 to 6: the same task description, evidence and options, then Your current choice:

\langle

answer of the previous round

\rangle

Your last message:

\langle

message of the previous round

\rangle

Messages from other participants: [1]

\langle

a peer’s message

\rangle

(choice:

\langle

its answer

\rangle

) standard message passing [1]

\langle

a restated evidence sentence the receiver lacks

\rangle

inhibitory message passing … Using their messages as additional information, reconsider your choice. Reason in at most 100 words, then end with a last line of the form "Final answer:

\langle

option

\rangle

Appendix

Table B.1: One prompt template for every condition. Angle brackets mark what each call fills in. Standard message passing fills the inbox with the sampled peers’ last messages and their answers, inhibitory message passing with the evidence sentences that peers’ messages restate and the receiver lacks, without answers; an empty inbox reads “(no new messages)”. On MuSiQue the agent gives an answer instead of an option. Two controls add one sentence each: share-all ( Table A.2 ) and declared tags ( Section C.4 ).

Figure C.1: Declared tags could serve as the router’s key on HiddenBench and MuSiQue but not on the puzzles. Development items. Left: group accuracy with the router keyed on declared tags or on word overlap, and with standard message passing shown the tags. Filled:

N=16

; open:

N=32

; MuSiQue only at

N=16

. Right: share of the holders’ first messages that declared their evidence sentence’s tag. 95% intervals.

Problem	In any system whose agents pass on what they hold, messages repeat the evidence every agent has, while an evidence sentence one agent holds reaches few receivers. The repetition grows with the group.
Solution	Each round, forward the evidence sentences each sender’s messages restate once to each receiver that lacks them, keyed on the source (here word overlap with the holder’s sentence; a source tag where available). Omit the senders’ reasoning and answers.
Consequences	Higher accuracy; 10–46% fewer tokens per run in peer groups; a shorter prompt for the decider; no model calls.
Applies to	Evidence that can be traced to original sentences, passed between agents in any architecture.
Tested on	Peer-to-peer, orchestrator–worker and hierarchical systems; 4 to 64 agents; nine models; evidence assigned to agents or found by their own searches (peer groups and a lead that delegates nothing); tasks with a scored answer.
Does not do	Check whether an evidence sentence is true. Combine the evidence for the receiver. Carry results agents derive, such as code or plans.

Appendix

Table C.1: Inhibitory message passing is a design pattern for routing evidence between agents. Token savings: the two main models, against standard message passing.

	Qwen3-8B	Olmo-3-7B
Peers: HiddenBench
over standard message passing	$+$ 75 [62, 86] ✓	$+$ 60 [50, 71] ✓
over the best prior-work baseline	$+$ 63 [52, 72] ✓	$+$ 44 [33, 54] ✓
every sender over six senders	$+$ 11 [5, 18] ✓	$+$ 34 [22, 47] ✓
late evidence (primacy index)	0.98 [ $-$ 1.05, 2.21] $\times$	0.61 [0.13, 0.96] ✓
Peers: Decoy puzzles

Appendix

Table D.1: Every pre-registered test passed on test items except the late-evidence test for Qwen3-8B on HiddenBench. Estimates with 95% intervals, after the exclusions. Rows under “Peers” give the router’s gain in group accuracy in the core comparison, in percentage points, over the named condition averaged over

N

(

N=4

–32; MuSiQue 4–16), or at the largest

N

over six senders. The late-evidence row, the primacy index, gives the share of benefit lost when private evidence arrives late, standard message passing minus the router. Architecture and fan-out/gather rows average over

N=8

, 16 and 32; in the architecture rows for peer-to-peer groups, under both communication rules each agent hears every peer, and the two conditions ran on different GPUs, so they do not share round 0. “All restated copies” is the router without cancellation ( Section H.1 ). Own-search rows are at

N=32

; a token ratio below 1 means fewer tokens. ✓/

\times

: passed/failed.

Benchmark	Model	Condition	$N=4$	$N=8$	$N=16$	$N=32$
HiddenBench	Ministral-3-8B	standard message passing	43	40	36	37
		inhibitory message passing	84	82	86	87
		full information	66	67	70	80
HiddenBench	Gemma-4-12B	standard message passing	55	39	34	28
		inhibitory message passing	95	93	96	96
		full information	91	94	97	95

Appendix

Table E.1: Ministral-3-8B, Gemma-4-12B and Qwen3 at 4B, 14B and 32B: group accuracy under the two conditions and with full information.

Model	Condition	$N=4$	$N=8$	$N=16$	$N=32$
GPT-6 Luna	standard message passing	39	31	27	18
	inhibitory message passing	94	93	92	91
	full information	94	96	96	95
	gain of the router [95% CI]	$+$ 55 [41, 68]	$+$ 61 [47, 74]	$+$ 65 [52, 76]	$+$ 74 [63, 83]
Claude Haiku 5.5	standard message passing	48	36	34	30
	inhibitory message passing	96	97	95	95

Appendix

Table E.2: Two commercial models on HiddenBench. Group accuracy per condition and

N

, and the router’s gain over standard message passing in percentage points with its 95% interval; 37 tasks for GPT-6 Luna and 36 for Claude Haiku 5.5 after the exclusions.

Qwen3-8B
		HiddenBench, $N=$			Decoy puzzles, $N=$
Architecture	Condition	8	16	32	8	16	32
Peer-to-peer	standard	7	4	2	11	12	12
	inhibitory	76	80	80	88	90	91
Fan-out/gather	standard	4	2	2	6	6	5
	all restated copies	43	30	20	22	20	19

Appendix

Table F.1: Inhibitory message passing raised group accuracy in every architecture and at every group size. Group accuracy over every run; Table D.1 applies the exclusions. Peer-to-peer groups hear every peer and vote; otherwise the decider answers.

Qwen3-8B
			HiddenBench, $N=$			Decoy puzzles, $N=$
Decider	Measure	Condition	8	16	32	8	16	32
Lead	private evidence reached	standard	62	68	68	67	70	70
		inhibitory	91	91	91	88	88	88
	shared-evidence copies	standard	13.5	27.0	58.3	18.4	38.4	76.0
		inhibitory	0.0	0.0	0.0	0.0	0.0	0.0

Appendix

Table F.2: Under the router, the decider’s inbox held no copy of the shared evidence and more private evidence, and its prompt was shorter. The percentage of private evidence sentences in the decider’s last inbox, and the copies of shared evidence sentences (summed over the shared evidence sentences) and the decider’s prompt tokens, averaged over rounds. Every run.

Figure F.1: Most private evidence sentences reached the decider under the router, yet its accuracy stayed below the share delivered. Share of private evidence in the decider’s last inbox against group accuracy, mean over

N=8

, 16 and 32. Filled: orchestrator–worker; open: hierarchical. Circles: Qwen3-8B; squares: Olmo-3-7B. Orange: standard message passing; blue: inhibitory message passing. Dashed diagonal: accuracy equal to the share delivered.

Figure G.1: When the agents’ own searches set the holders, inhibitory message passing raised group accuracy at fewer tokens. MuSiQue,

N=32

. ( a ) Group accuracy of peer groups and of a fan-out/gather lead; Table D.1 gives the intervals. ( b ) Tokens per run relative to standard message passing, with 95% intervals.

Variant	Drops	What the receiver gets
LLM-extracted claims	(1)	the factual claims an LLM lists from each sender’s message, without its opinions or conclusions
sender’s sentence	(1)	the sentence of the sender’s latest message that restated each routed evidence sentence
sender’s sentence, all copies	(1), (2), (3)	those sentences from every sender, without cancellation
LLM paraphrase of the evidence	nothing; the text reworded	each routed evidence sentence reworded once by the same model
latest message only	nothing; the record shortened	only the sentences restated in the sender’s latest message, not all it has restated
+ answers only	nothing; answers added	the routed sentences and each sender’s answer

Appendix

Table H.1: The variants of the router. Each keeps the router’s matching and its sender set and changes one thing. The rules are those of Equation 1b : (1) evidence, not messages; (2) once per round; (3) only if missing.

Condition	HB $N{=}16$	HB $N{=}32$	puzzles strong	puzzles weak	MQ $N{=}16$
no communication	2 / 11	1 / 9	0 / 0	16 / 7	5 / 1
standard message passing	8 / 21	5 / 21	0 / 1	19 / 15	12 / 9
standard message passing with every sender †	2 / 23	3 / 20	1 / 5	17 / 14	–
inhibitory message passing, 6 senders	88 / 81	79 / 57	45 / 36	74 / 52	52 / 42
inhibitory message passing	88 / 88	91 / 92	87 / 76	84 / 74	57 / 55
+ answers only †	84 / 90	84 / 94	81 / 62	67 / 57	42 / 30

Appendix

Table H.2: Taking the router apart on development items (group accuracy, Qwen3-8B / Olmo-3-7B; HB: HiddenBench; puzzles by decoy strength and MQ (MuSiQue) at

N=16

). The third row is standard message passing hearing every peer ( Section 3 ); the indented rows are variants of the router ( Table H.1 ). † : exploratory, run after the design was fixed.

Figure H.1: Accuracy fell as more copies of each shared evidence sentence reached a receiver, most clearly on the puzzles. Development items, exploratory. Group accuracy against the copies of each shared evidence sentence per inbox that pass the router, without cancellation as the group grows (gray;

N=4

to 32) and with

k=1

, 2 or 4 copies let through at

N=16

(blue;

k=0

is the router). The x axis is logarithmic in 1 plus the copies. 95% intervals.

group accuracy
	Qwen3-8B				Olmo-3-7B
	HiddenBench		puzzles		HiddenBench		puzzles
Condition	$N=16$	$N=32$	$N=16$	$N=32$	$N=16$	$N=32$	$N=16$	$N=32$
standard message passing	6	7	11	10	16	10	10	10
+ full inbox history	7	5	12	12	19	22	10	10
all restated copies	53	42	24	20	48	44	8	7

Appendix

Table H.3: Simpler filters : group accuracy (%) and tokens per run of the random sample, deduplication only and the LLM summary of messages, which keep the router’s inbox size, and of “+ full inbox history”, all restated copies and deduplication only, no size limit, which do not; Qwen3-8B and Olmo-3-7B on HiddenBench and the puzzles at

N=16

and 32. Levels are per condition after its own exclusions; the differences quoted in the text are paired on the items both conditions keep, so they can differ from the difference of two levels by a few points.

Evidence sentences restated (what inhibitory message passing forwards)	The sender’s sentence that restated them (what the sender’s sentence forwards)
Station Delta is the oldest building and sometimes has plumbing issues, but is currently empty. Maintenance crew at Station Delta inspected all major systems this morning and confirmed water is safe, backup lights work, and no signs of contamination were found.	Delta offers reliable power, safe water, and is currently empty—making it the safest and most functional option.
The Community Center basement has had a chemical leak that was just discovered today; the evacuation was for this, and it’s only safe for short visits, not day-long events.	The Community Center is now confirmed safe for a full day, with no chemical leak impact on main areas.
The City Library has a large conference room and claims to be ready; it also has a known backup generator for power. Library staff admit privately that their backup generator only has enough fuel for 2 hours and cannot be refueled due to a citywide gasoline shortage.	The Library’s generator only lasts 2 hours, insufficient for the full day.
Due to minor flooding on Oakcrest road, buses can only reach Riverside Park from the north side.	Although buses can only reach Riverside from the north due to Oakcrest road flooding, that route is clear.

Appendix

Table H.4: What the sender’s sentence delivers in place of the original evidence sentence : four (evidence sentence, restating sentence) pairs drawn at random (Qwen3-8B, HiddenBench development,

N=16

). They illustrate a failure mode, not its frequency.

Figure I.1: When the needed evidence grows with the group, delivery and integration fall behind. Elimination problems, development items,

H=N/4

: group accuracy under standard message passing, all restated copies, inhibitory message passing and full information, with 95% intervals. Blue numbers: missing evidence sentences the router delivered per inbox, Qwen3-8B. Solid lines and circles: Qwen3-8B; dashed lines and squares: Olmo-3-7B.

	Qwen3-8B		Olmo-3-7B
Condition	reference $\to$ planted	change	reference $\to$ planted	change
no communication	2 $\to$ 3	$+$ 2 [0, 4]	11 $\to$ 11	0 [0, 0]
standard message passing	8 $\to$ 5	$-$ 3 [ $-$ 6, 0]	21 $\to$ 23	$+$ 2 [ $-$ 7, 9]
inhibitory message passing, 6 senders	88 $\to$ 78	$-$ 10 [ $-$ 18, $-$ 3]	81 $\to$ 66	$-$ 15 [ $-$ 24, $-$ 8]
inhibitory message passing	88 $\to$ 82	$-$ 6 [ $-$ 16, 4]	88 $\to$ 79	$-$ 8 [ $-$ 16, $-$ 2]
full information, one answer	81 $\to$ 47	$-$ 34 [ $-$ 47, $-$ 21]	74 $\to$ 34	$-$ 41 [ $-$ 56, $-$ 26]

Appendix

Table I.1: A planted falsehood (HiddenBench development,

N=16

): group accuracy without

\to

with the false sentence, and the change in percentage points (95% interval).

Circuit-model concept	Router counterpart
neuron or local circuit	receiving agent
shared feedforward input	evidence sentence initially held by multiple agents
input to a subset of neurons	evidence sentence initially held by only some agents
recurrent excitation	reinforcing influence of peers’ reasoning and answers
excitation–inhibition balance input by input (detailed balance)	cancellation computed per evidence sentence and receiver
balance of total excitation and inhibition (global balance)	limit on total inbox size, independent of content

Appendix

Table J.1: Computational analogy between cortical circuit models and evidence routing.

Operation	Precedent	What it leaves open
forward verbatim evidence	debaters with a quote tool; quotes help the debater of the correct answer ( Khan et al., 2024 )	two debaters and a judge; no group of deciders, no scaling
drop the answers	an anonymous evidence board read by one judge ( Ji et al., 2026 ) ; removing self and peer labels curbs identity-driven sycophancy ( Choi et al., 2026 )	one judge, or labels removed while answers still circulate; not tested with distributed evidence
subtract what the receiver holds	a receiver discards peers’ answers similar to its own or earlier ones ( Zeng et al., 2025 ) ; learned pruning of edges ( Zhang et al., 2025a ) ; per-receiver merging of similar messages ( Guan et al., 2026 ) ; sensor nodes advertise data so that each receives only what it lacks ( Heinzelman et al., 1999 )	LLM methods filter on answers, similarity or learned value, for cost; SPIN (sensor protocols for information via negotiation) moves sensor data, not evidence among reasoning agents
pool	blackboards ( Erman et al., 1980 ; Han and Zhang, 2025 )	every receiver reads the whole board; a redundant entry is removed for all receivers, not per receiver
count each piece of evidence once	the probabilistic norm and the channel filter of decentralized fusion ( Pearl, 1988 ; Grime and Durrant-Whyte, 1994 ) ; incest removal ( Krishnamurthy and Hoiles, 2014 ) ; “data incest” among LLM agents ( Jain and Krishnamurthy, 2025 ) ; people neglect correlated sources ( Enke and Zimmermann, 2019 ) ; LLM aggregators count restated copies as partial new evidence ( Gao et al., 2026 )	the fusion filters are exact on trees only; among LLM agents, measured at one aggregator and corrected by prompting in four-agent teams; no communication rule removes copies before every receiver
brain-inspired communication	agents as cortical areas in a sparse, trust-weighted debate graph ( Sun et al., 2025 )	topology, not what a message carries

Appendix

Table K.1: Precedents of each operation of inhibitory message passing.

Multi-Agent Reasoning

Multi-Agent Reasoning studies how multiple AI agents share evidence, model one another, and coordinate decisions on complex tasks.

See all 232 papers

Oct 14, 2025

Benefits and Limitations of Communication in Multi-Agent Reasoning

Chain-of-thought prompting has popularized step-by-step reasoning in large language models, yet model performance still degrades as problem complexity and context length grow. By decomposing difficult tasks with long contexts into shorter, manageable ones, recent multi-agent paradigms offer a promising near-term solution to this problem. However, the fundamental capacities of such systems are poorly understood. In this work, we propose a theoretical framework to analyze the expressivity of multi-agent systems. We apply our framework to three algorithmic families: state tracking, recall, and

k

-hop reasoning. We derive bounds on (i) the number of agents required to solve the task exactly, (ii) the quantity and structure of inter-agent communication, and (iii) the achievable speedups as problem size and context scale. Our results identify regimes where communication is provably beneficial, delineate tradeoffs between agent count and bandwidth, and expose intrinsic limitations when either resource is constrained. We complement our theoretical analysis with a set of experiments on pretrained LLMs using controlled synthetic benchmarks. Empirical outcomes confirm the tradeoffs between key quantities predicted by our theory. Collectively, our analysis offers principled guidance for designing scalable multi-agent reasoning systems.

Michael Rizvi-Martel, Satwik Bhattamishra, Neil Rathi +2

Mila & Universit´e de Montr´eal · University of Oxford · Stanford University +1

Multi-Agent Reasoning Communication Complexity Multi-Agent Communication Multi-Agent LLMs

Ask PDF

Oct 26, 2026

Multi-Agent Relay Thinking

Existing multi-agent collaboration (MAC) methods have primarily explored a group thinking collaboration topology, where agents independently generate solutions and aggregate their outputs. In this work, we investigate an alternative relay thinking topology, where reasoning ownership is dynamically transferred across agents, allowing subsequent agents to inherit and refine an evolving reasoning trajectory. We introduce Multi-Agent Confidence-guided Relay (MACR), a framework that enables dynamical relay collaboration among agents. The key challenge is determining when reasoning should be relayed and which agent should continue the process. method addresses this challenge through confidence-guided relay triggering and a bidding mechanism that selects the next agent based on its confidence. Experiments on four benchmarks show that method outperforms the strongest single-model strategies in accuracy, e.g., +3.3% on MathVista and +3.9% on MMStar, and remains on par with multi-agent strategies while consuming 6.0x less token overhead on HallusionBench and MMStar.Further analysis demonstrates that relay collaboration enables heterogeneous agents to progressively integrate complementary insights, combining heterogeneous capabilities across reasoning stages. Our source code is available at https://github.com/67L1/MACR.

Xu Liu, Jungang Qiao, Guikun Chen +2

Zhejiang University · Nanyang Technological University · The University of Hong Kong

Multi-Agent LLM Systems Multi-Agent Collaboration Multi-Agent Reasoning Multi-Agent Deliberation

Ask PDF

Sep 28, 2026

Self-Adapting Group of Experts for Multi-Agent Reasoning

Multi-agent systems bring together language model agents with different roles to propose, review, and refine solutions. Each agent's response depends on its model's capabilities, the reasoning strategy defined by its system prompt, and the information in its input context. Existing frameworks often adapt communication by changing this context while leaving individual prompts fixed, even when a problem calls for different skills. We study whether agents' initial responses can identify a strategy better suited to the current problem and guide its transfer to other agents. To address this, we introduce SAGE (Self-Adapting Group of Experts), a training-free framework that uses answer agreement, prefix consistency, and reciprocal peer review to select a strategy donor. SAGE transfers the selected donor's reasoning strategy to the other agents while preserving their original roles. This transfer uses only the agents' original system prompts, without access to the problem or generated solutions. After strategy adaptation, agents exchange responses through a dynamic, sparse directed acyclic graph that routes information from higher-scoring agents to lower-scoring agents. Experiments across multiple agent backbones and reasoning benchmarks show that SAGE achieves higher average accuracy than the evaluated baselines. Our code is available at https://github.com/atifquamar07/sage.

Mohammad Atif Quamar, Nurbek Tastan, Karthik Nandakumar +1

Mohamed bin Zayed University of Artificial Intelligence · Michigan State University · RIKEN AIP

Multi-Agent LLM Systems Multi-Agent Reasoning Multi-Agent Communication

Ask PDF

Abstract

Multi-Agent Reasoning

Explore similar work

Benefits and Limitations of Communication in Multi-Agent Reasoning

Multi-Agent Relay Thinking

Self-Adapting Group of Experts for Multi-Agent Reasoning