J-Comm: Jacobian-Guided Latent Communication between LLM Agents

Xiangwei Wang; Wei Wang; Sachith Seneviratne; Ken Chen; Yu Xia; Peng Wang; Saman Halgamuge

cs.CLOct 23, 2026

J-Comm: Jacobian-Guided Latent Communication between LLM Agents

Authors: Xiangwei Wang, Wei Wang, Sachith Seneviratne, Ken Chen, Yu Xia, Peng Wang, Saman Halgamuge

Organizations: The University of Melbourne · Shanghai Jiao Tong University

Abstract

Latent communication requires both an effective continuous reasoning process and an interface through which an external model can leverage its intermediate computation. We present J-Comm, a framework comprising two core components. First, a training-free Geometric Agreement for Latent Reasoning (GEAR) reweights token candidates over existing model embeddings, which monotonically optimizes geometric consensus among candidate tokens in embedding space at each step. Second, we introduce an H+J (Hidden-and-Jacobian) communication interface that gives the Receiver two complementary views of the Sender's reasoning. The H channel supplies continuous inputs derived from feedback distributions, while the J channel conveys intermediate-state cues in vocabulary-indexed Jacobian-lens coordinates. Intuitively, H provides inputs for the Receiver to reason from, while J offers additional guidance within its computation. The Receiver reconstructs these cues using its own dictionary and injects them into intermediate layers during prefill, enabling communication without sharing key-value caches or requiring identical model widths. In Qwen3-8B to Qwen3-4B communication, J-Comm achieves the highest mean score across seven benchmarks, including 94.80% on MATH500, 54.55% on GPQA-Diamond, 73.33% on AIME2024+2025, and 62.93% success on ALFWorld. Compared with TextMAS, it improves performance by 1.29-8.59 percentage points and reduces generated text tokens by 19.2-74.8%.

Figures & tables

Figure 1: Overview of J-Comm . GEAR reweights candidate probabilities by geometric agreement and feeds the resulting embedding mixture into the Sender’s next position. The final $M$ consumed steps supply two messages: $H$ maps their feedback distributions into Receiver input embeddings, while $J$ carries token-anchor IDs and signed intermediate-state readouts for reconstruction in the Receiver’s own dictionary. The $J$ update is added at the $M$ message positions during prefill. Top-P denotes a fixed number of anchors ( $P=32$ ), distinct from nucleus sampling. The two displayed blocks use index 26 in our main setting.

Method	GSM8K	MATH500	GPQA-Diamond	MedQA	MBPP+	AIME24+25	ALFWorld
Text MAS	93.25 (ref.)	93.20 (ref.)	45.96 (ref.)	78.55 (ref.)	71.16 (ref.)	68.33 (ref.)	61.39 (ref.)
Receiver only	86.58 ( $\downarrow$ 36.5%)	88.20 ( $\downarrow$ 32.6%)	39.90 ( $\downarrow$ 22.5%)	69.76 ( $\downarrow$ 20.8%)	72.22 ( $\downarrow$ 27.6%)	67.78 ( $\uparrow$ 0.6%)	49.25 ( $\downarrow$ 55.8%)
Identity40+ $H+J26$	86.73 ( $\downarrow$ 36.6%)	89.40 ( $\downarrow$ 32.0%)	39.39 ( $\downarrow$ 22.5%)	70.23 ( $\downarrow$ 25.4%)	71.16 ( $\downarrow$ 26.7%)	69.44 ( $\uparrow$ 9.1%)	53.73 ( $\downarrow$ 56.4%)
Identity( $T_{\rm GAF}$ )+ $H$	87.19 ( $\downarrow$ 38.7%)	89.00 ( $\downarrow$ 34.4%)	39.39 ( $\downarrow$ 27.0%)	71.25 ( $\downarrow$ 26.9%)	71.69 ( $\downarrow$ 29.5%)	65.00 ( $\uparrow$ 1.4%)	54.25 ( $\downarrow$ 58.1%)
Identity( $T_{\rm GAF}$ )+ $H+J26$	87.41 ( $\downarrow$ 38.8%)	89.00 ( $\downarrow$ 35.3%)	43.43 ( $\downarrow$ 26.0%)	70.15 ( $\downarrow$ 27.5%)	73.02 ( $\downarrow$ 31.0%)	71.11 ( $\downarrow$ 3.6%)	57.32 ( $\downarrow$ 60.3%)
GAF+ $H$	92.12 ( $\downarrow$ 47.5%)	93.60 ( $\downarrow$ 54.5%)	49.49 ( $\downarrow$ 48.4%)	78.40 ( $\downarrow$ 54.1%)	73.54 ( $\downarrow$ 35.9%)	72.78 ( $\downarrow$ 8.0%)	59.70 ( $\downarrow$ 66.2%)

Table 1: Qwen3-8B

\to

Qwen3-4B results averaged over three runs. Cells report accuracy, pass@1, or ALFWorld success rate (%), followed by the percentage change in generated text tokens relative to TextMAS (

\downarrow

: fewer tokens;

\uparrow

: more tokens). TextMAS includes Sender thinking and plan generation plus Receiver output; latent methods generate text at the Receiver. ALFWorld token totals include all decisions in each task. Bold marks the highest mean score in each column.

Figure 2: Mean GAF accuracy gains over matched-step Soft Thinking and Self-only across three runs on MATH500 and GPQA-Diamond, with a fixed

H+J26

interface. Error bars show question-paired 95% bootstrap intervals.

Figure 3: Receiver-layer ablation on AIME2024+2025 with the same 400-position

H

message. Points and error bars show mean

\pm

SD across three runs; the dashed line and band show the

H

-only mean

\pm

SD. The

H

-only and J26 results also appear in Table 1 .

Figure 4: Communication-length ablation on AIME2024+2025 using GAF+

H+J26

and a shared Sender cache. Points and error bars show mean

\pm

SD across three runs; the dashed line and band show the TextMAS mean

\pm

SD from Table 1 .

	Qwen3-4B $\to$ 4B			Qwen3-8B $\to$ 8B
Method	MATH500	GPQA-Diamond	AIME24+25	MATH500	GPQA-Diamond	AIME24+25
GAF+ $H+J26$	92.00	41.92	68.33	95.20	49.49	75.00
LatentMAS	83.20	38.38	61.67	86.80	47.10	51.67
StateBridge	89.40	45.45	61.67	93.20	52.50	68.33
CIPHER	80.40	35.35	58.33	84.20	41.41	61.67

Table 2: Homogeneous two-agent communication accuracy (%), averaged over three runs. GAF+

H+J26

uses 400 communicated positions.

Appendix figures & tables4 assets

Supplementary material from the paper’s appendix.

Appendix

Setting	Value
Sender / Receiver	Qwen3-8B / Qwen3-4B
Sender / Receiver block	26 / 26
Sender temperature	0.6
Initial top- $k$ candidates	30
Nucleus threshold	0.95
Relative min- $p$ threshold	0.001

Appendix

Table 3: Default Qwen3-8B

\to

Qwen3-4B configuration. Sender latent steps, communication positions, and Receiver output tokens have separate budgets.

Dataset	Sender token cap	Receiver token cap
GSM8K	1,024	2,048
MATH500	4,096	8,192
GPQA-Diamond	4,096	8,192
MedQA	2,048	4,096
MBPP+	2,048	4,096
AIME2024+2025	8,192	31,000

Appendix

Table 4: Maximum text output tokens per agent. ALFWorld budgets apply to each action turn, with at most 50 turns per episode.

Task	R	Text	$H$	$HJ$	$HJ_{\rm shuf}$
GPQA	40	73	71	75	69

Appendix

Table 5: Qwen3-8B

\to

Llama-3.2-3B-Instruct on GPQA-Diamond: mean correct answers over three runs, out of 198 questions. Each condition uses 40 communicated positions when applicable and an 8,192-token Receiver cap; the TextMAS Sender cap is 4,096.

HJ

writes at native Llama block 20.

Figure 5: Inspectable $J$ coordinates and shuffled donors for a MedQA apoptosis question (a–d) and MATH500 square-table counting (e–h), both with 400 communication positions. Gray cells denote absent sparse coordinates. All three medical-case conditions answer correctly; only matched $J$ completes the counting solution within the output budget.

Multi-Agent Communication

Multi-agent communication studies how autonomous agents exchange information to coordinate decisions, share knowledge, and act together under constraints.

See all 380 papers

Beyond tokens: a unified framework for latent communication in LLM-based multi-agent systems
Jun 4, 2026Yingzhuo LiuMulti-Agent LLM Systems Multi-Agent Communication
LatComm: Reusable Context Compression for Latent Communication between Large Language Models
Oct 19, 2026Huawei Lin, Jinting Liu, Tan Yu +1LLM Context Compression LLM Inference Acceleration
Rochester Institute of Technology · NVIDIA
StateBridge: Training-free Hidden-state Alignment for Latent Communication in LLM Multi-Agent Systems
Aug 13, 2026Yanwen Peng, Delvin Ce Zhang, Xi Wang +1Multi-Agent LLM Systems Multi-Agent Communication
School of Computer Science, University of Sheffield Sheffield, United Kingdom

Jun 4, 2026J/K move · Enter open · S save

Beyond tokens: a unified framework for latent communication in LLM-based multi-agent systems

Yingzhuo Liu

Multi-Agent LLM Systems Multi-Agent Communication

Open paper AskPDF arXiv

Multi-agent systems built on large language models (LLMs) have become a prevailing paradigm for tackling complex reasoning, planning, and tool-use tasks. The dominant communication protocol in such systems is natural language: agents exchange messages token-by-token, verbalising their internal reasoning so that peers can read, verify, and respond. While convenient and interpretable, this protocol suffers from three structural drawbacks -- high inference cost, irreversible information loss during discretization, and ambiguity/redundancy of natural language. A growing body of work therefore explores an alternative protocol -- latent communication -- in which agents exchange continuous representations (embeddings, hidden states, or KV-caches) directly, bypassing the bottleneck of text generation. This paper presents a unified framework for organising the rapidly expanding literature on latent communication. We analyse existing methods along three orthogonal axes: (1) WHAT information is communicated (Embeddings, Hidden States, KV-Caches, or other continuous state); (2) WHICH sender-receiver alignment is used (latent-space alignment and layer alignment); and (3) HOW the communicated information is fused into the receiver (concatenation, prepending, mathematical operations, cross-attention, or cache restoration). Under this 3-axis framework, we systematically categorise eighteen representative methods proposed between 2024 and 2026, identify five major design patterns, and surface a set of open challenges -- including cross-architecture alignment, security of latent channels, compression for edge deployment, and the relationship between latent communication and latent chain-of-thought. We hope that this framework both lowers the barrier to entry for new researchers and provides a vocabulary for comparing future work.

Abstract

Multi-Agent Communication

Explore similar work

Beyond tokens: a unified framework for latent communication in LLM-based multi-agent systems

LatComm: Reusable Context Compression for Latent Communication between Large Language Models

StateBridge: Training-free Hidden-state Alignment for Latent Communication in LLM Multi-Agent Systems