@inproceedings{tang-etal-2026-halt,
title = "{HALT}: Evaluator-Guided Clause-Level Pruning and Early Halting for Lower-Cost Multi-Agent Systems",
author = "Tang, Zongwei and
Li, Yewen and
Liu, Bo and
Zhang, Zezong and
Yiqing, Cai and
Guo, Yiwei and
Qiu, Houjie and
Ma, Xingkong",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1012/",
pages = "23283--23307",
ISBN = "979-8-89176-459-0",
abstract = "Multi-agent systems enhance complex reasoning but face a critical cost and effectiveness dilemma: API costs grow exponentially while performance degrades due to context degradation and error propagation. This paper proposes an evaluator-guided clause-level pruning and early halting method for lower-cost multi-agent systems named HALT, which implements a macro-to-micro intervention. At the macro-level, a difficulty aware router classifies query complexity, bypassing redundant multi-agent systems overhead for simple tasks. At the micro-level, a lightweight hindsight evaluator monitors generation in real time for complex reasoning, driving a dual-axis intervention policy: spatial pruning removes erroneous clauses, while temporal halting terminates multi-agent interaction early upon reaching a purified consensus. The extensive experiments across three different datasets demonstrate the outperformance of HALT over the baselines, achieving the 97.98{\%}, 90.82{\%}, 87.14{\%} in GSM8K, MMLU, MMLU-Pro in accuracy while drastically mitigating computational overhead, with a maximum reduction of 94.3{\%} in prompt tokens and an unprecedented reduction of 98.9{\%} in completion tokens, which validate the effectiveness of HALT. Code is available at URL."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="tang-etal-2026-halt">
<titleInfo>
<title>HALT: Evaluator-Guided Clause-Level Pruning and Early Halting for Lower-Cost Multi-Agent Systems</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zongwei</namePart>
<namePart type="family">Tang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yewen</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bo</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zezong</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Cai</namePart>
<namePart type="family">Yiqing</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yiwei</namePart>
<namePart type="family">Guo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Houjie</namePart>
<namePart type="family">Qiu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xingkong</namePart>
<namePart type="family">Ma</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Multi-agent systems enhance complex reasoning but face a critical cost and effectiveness dilemma: API costs grow exponentially while performance degrades due to context degradation and error propagation. This paper proposes an evaluator-guided clause-level pruning and early halting method for lower-cost multi-agent systems named HALT, which implements a macro-to-micro intervention. At the macro-level, a difficulty aware router classifies query complexity, bypassing redundant multi-agent systems overhead for simple tasks. At the micro-level, a lightweight hindsight evaluator monitors generation in real time for complex reasoning, driving a dual-axis intervention policy: spatial pruning removes erroneous clauses, while temporal halting terminates multi-agent interaction early upon reaching a purified consensus. The extensive experiments across three different datasets demonstrate the outperformance of HALT over the baselines, achieving the 97.98%, 90.82%, 87.14% in GSM8K, MMLU, MMLU-Pro in accuracy while drastically mitigating computational overhead, with a maximum reduction of 94.3% in prompt tokens and an unprecedented reduction of 98.9% in completion tokens, which validate the effectiveness of HALT. Code is available at URL.</abstract>
<identifier type="citekey">tang-etal-2026-halt</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1012/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>23283</start>
<end>23307</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T HALT: Evaluator-Guided Clause-Level Pruning and Early Halting for Lower-Cost Multi-Agent Systems
%A Tang, Zongwei
%A Li, Yewen
%A Liu, Bo
%A Zhang, Zezong
%A Yiqing, Cai
%A Guo, Yiwei
%A Qiu, Houjie
%A Ma, Xingkong
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F tang-etal-2026-halt
%X Multi-agent systems enhance complex reasoning but face a critical cost and effectiveness dilemma: API costs grow exponentially while performance degrades due to context degradation and error propagation. This paper proposes an evaluator-guided clause-level pruning and early halting method for lower-cost multi-agent systems named HALT, which implements a macro-to-micro intervention. At the macro-level, a difficulty aware router classifies query complexity, bypassing redundant multi-agent systems overhead for simple tasks. At the micro-level, a lightweight hindsight evaluator monitors generation in real time for complex reasoning, driving a dual-axis intervention policy: spatial pruning removes erroneous clauses, while temporal halting terminates multi-agent interaction early upon reaching a purified consensus. The extensive experiments across three different datasets demonstrate the outperformance of HALT over the baselines, achieving the 97.98%, 90.82%, 87.14% in GSM8K, MMLU, MMLU-Pro in accuracy while drastically mitigating computational overhead, with a maximum reduction of 94.3% in prompt tokens and an unprecedented reduction of 98.9% in completion tokens, which validate the effectiveness of HALT. Code is available at URL.
%U https://aclanthology.org/2026.emnlp-main.1012/
%P 23283-23307
Markdown (Informal)
[HALT: Evaluator-Guided Clause-Level Pruning and Early Halting for Lower-Cost Multi-Agent Systems](https://aclanthology.org/2026.emnlp-main.1012/) (Tang et al., EMNLP 2026)
ACL
- Zongwei Tang, Yewen Li, Bo Liu, Zezong Zhang, Cai Yiqing, Yiwei Guo, Houjie Qiu, and Xingkong Ma. 2026. HALT: Evaluator-Guided Clause-Level Pruning and Early Halting for Lower-Cost Multi-Agent Systems. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 23283–23307, Budapest, Hungary. Association for Computational Linguistics.