More with Less: An Empirical Study of Turn-Control Strategies for Efficient Coding Agents

<p dir="ltr">LLM-powered coding agents, which operate in iterative loops (turns) to solve software engineering tasks, are becoming increasingly powerful. However, their practical deployment is hindered by significant and unpredictable costs. This challenge arises from a combination o...

Cur síos iomlán

Sábháilte in:
Sonraí bibleagrafaíochta
Príomhchruthaitheoir: Chao Peng (22681619) (author)
Rannpháirtithe: Pengfei Gao (13470673) (author)
Foilsithe / Cruthaithe: 2025
Ábhair:
Clibeanna: Cuir clib leis
Níl clibeanna ann, Bí ar an gcéad duine le clib a chur leis an taifead seo!
_version_ 1849927634633359360
author Chao Peng (22681619)
author2 Pengfei Gao (13470673)
author2_role author
author_facet Chao Peng (22681619)
Pengfei Gao (13470673)
author_role author
dc.creator.none.fl_str_mv Chao Peng (22681619)
Pengfei Gao (13470673)
dc.date.none.fl_str_mv 2025-11-25T12:25:13Z
dc.identifier.none.fl_str_mv 10.6084/m9.figshare.30705524.v1
dc.relation.none.fl_str_mv https://figshare.com/articles/dataset/More_with_Less_An_Empirical_Study_of_Turn-Control_Strategies_for_Efficient_Coding_Agents/30705524
dc.rights.none.fl_str_mv CC BY 4.0
info:eu-repo/semantics/openAccess
dc.subject.none.fl_str_mv Automated software engineering
Empirical software engineering
LLM
Agent
dc.title.none.fl_str_mv More with Less: An Empirical Study of Turn-Control Strategies for Efficient Coding Agents
dc.type.none.fl_str_mv Dataset
info:eu-repo/semantics/publishedVersion
dataset
description <p dir="ltr">LLM-powered coding agents, which operate in iterative loops (turns) to solve software engineering tasks, are becoming increasingly powerful. However, their practical deployment is hindered by significant and unpredictable costs. This challenge arises from a combination of factors: quadratically growing token counts with each turn, the high price of state-of-the-art models, the large number of turns required for real-world tasks, and the tendency of agents to take inefficient or unnecessary actions. While existing research focuses on optimizing individual turns, the strategic control of the total number of turns remains an underexplored area for managing agent performance and cost. To address this gap, we conduct a comprehensive empirical study on the SWE-bench benchmark using three state-of-the-art models (Claude 4 Sonnet, Gemini 2.5 Pro, and GPT 4.1). We systematically evaluate the impact of three distinct turn-control strategies: an unrestricted baseline, a fixed-turn limit with reminders, and a novel dynamic-turn strategy that grants extensions on-demand. Our findings first reveal a fundamental trade-off in the unrestricted setting, where no single model excels across performance, cost, and turn efficiency. We then show that a fixed-turn limit, specifically at the 75th percentile of the baseline, serves as a "sweet spot", substantially reducing costs (by 24%-68%) with minimal impact on solve rates. Most significantly, our proposed dynamic-turn strategy consistently outperforms fixed-limit approaches, achieving comparable or better solve rates while further reducing costs by an additional 12%-24% by intelligently allocating resources only to tasks that need them. This work provides the first systematic analysis of turn-control strategies, offering simple yet effective guidelines for developers to balance cost and efficacy. We demonstrate that dynamic resource allocation is a superior, easy-to-implement approach for deploying powerful yet economically viable coding agents.</p>
eu_rights_str_mv openAccess
id Manara_f5a1121dee809444208d5c918e971f06
identifier_str_mv 10.6084/m9.figshare.30705524.v1
network_acronym_str Manara
network_name_str ManaraRepo
oai_identifier_str oai:figshare.com:article/30705524
publishDate 2025
repository.mail.fl_str_mv
repository.name.fl_str_mv
repository_id_str
rights_invalid_str_mv CC BY 4.0
spelling More with Less: An Empirical Study of Turn-Control Strategies for Efficient Coding AgentsChao Peng (22681619)Pengfei Gao (13470673)Automated software engineeringEmpirical software engineeringLLMAgent<p dir="ltr">LLM-powered coding agents, which operate in iterative loops (turns) to solve software engineering tasks, are becoming increasingly powerful. However, their practical deployment is hindered by significant and unpredictable costs. This challenge arises from a combination of factors: quadratically growing token counts with each turn, the high price of state-of-the-art models, the large number of turns required for real-world tasks, and the tendency of agents to take inefficient or unnecessary actions. While existing research focuses on optimizing individual turns, the strategic control of the total number of turns remains an underexplored area for managing agent performance and cost. To address this gap, we conduct a comprehensive empirical study on the SWE-bench benchmark using three state-of-the-art models (Claude 4 Sonnet, Gemini 2.5 Pro, and GPT 4.1). We systematically evaluate the impact of three distinct turn-control strategies: an unrestricted baseline, a fixed-turn limit with reminders, and a novel dynamic-turn strategy that grants extensions on-demand. Our findings first reveal a fundamental trade-off in the unrestricted setting, where no single model excels across performance, cost, and turn efficiency. We then show that a fixed-turn limit, specifically at the 75th percentile of the baseline, serves as a "sweet spot", substantially reducing costs (by 24%-68%) with minimal impact on solve rates. Most significantly, our proposed dynamic-turn strategy consistently outperforms fixed-limit approaches, achieving comparable or better solve rates while further reducing costs by an additional 12%-24% by intelligently allocating resources only to tasks that need them. This work provides the first systematic analysis of turn-control strategies, offering simple yet effective guidelines for developers to balance cost and efficacy. We demonstrate that dynamic resource allocation is a superior, easy-to-implement approach for deploying powerful yet economically viable coding agents.</p>2025-11-25T12:25:13ZDatasetinfo:eu-repo/semantics/publishedVersiondataset10.6084/m9.figshare.30705524.v1https://figshare.com/articles/dataset/More_with_Less_An_Empirical_Study_of_Turn-Control_Strategies_for_Efficient_Coding_Agents/30705524CC BY 4.0info:eu-repo/semantics/openAccessoai:figshare.com:article/307055242025-11-25T12:25:13Z
spellingShingle More with Less: An Empirical Study of Turn-Control Strategies for Efficient Coding Agents
Chao Peng (22681619)
Automated software engineering
Empirical software engineering
LLM
Agent
status_str publishedVersion
title More with Less: An Empirical Study of Turn-Control Strategies for Efficient Coding Agents
title_full More with Less: An Empirical Study of Turn-Control Strategies for Efficient Coding Agents
title_fullStr More with Less: An Empirical Study of Turn-Control Strategies for Efficient Coding Agents
title_full_unstemmed More with Less: An Empirical Study of Turn-Control Strategies for Efficient Coding Agents
title_short More with Less: An Empirical Study of Turn-Control Strategies for Efficient Coding Agents
title_sort More with Less: An Empirical Study of Turn-Control Strategies for Efficient Coding Agents
topic Automated software engineering
Empirical software engineering
LLM
Agent