Saved in:
Bibliographic Details
Main Authors: CodeArts Model Team, Ye, Yang, Tan, Jingyuan, Jiang, Tianyue, Ye, Ruizhe, He, Qiankun, Yang, Jiarui, Dong, Jian, Liang, Sicong, Yue, Chongjian, Xu, Peibai, Lu, Lufan, Pang, Shiguan, Qian, Taotao, Hu, Junbao, Hao, Yuechan, Shi, Ensheng, Zhang, Qi, Hao, Yi, Fan, Na, Tan, Xin, Yao, Shuai, Shen, Zhiwei, Li, Zongchen, Wang, Yanlin, Chen, Chong, Ma, Yuchi
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.00824
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917385335734272
author CodeArts Model Team
Ye, Yang
Tan, Jingyuan
Jiang, Tianyue
Ye, Ruizhe
He, Qiankun
Yang, Jiarui
Dong, Jian
Liang, Sicong
Yue, Chongjian
Xu, Peibai
Lu, Lufan
Pang, Shiguan
Qian, Taotao
Hu, Junbao
Hao, Yuechan
Shi, Ensheng
Zhang, Qi
Hao, Yi
Fan, Na
Tan, Xin
Yao, Shuai
Shen, Zhiwei
Li, Zongchen
Wang, Yanlin
Chen, Chong
Ma, Yuchi
author_facet CodeArts Model Team
Ye, Yang
Tan, Jingyuan
Jiang, Tianyue
Ye, Ruizhe
He, Qiankun
Yang, Jiarui
Dong, Jian
Liang, Sicong
Yue, Chongjian
Xu, Peibai
Lu, Lufan
Pang, Shiguan
Qian, Taotao
Hu, Junbao
Hao, Yuechan
Shi, Ensheng
Zhang, Qi
Hao, Yi
Fan, Na
Tan, Xin
Yao, Shuai
Shen, Zhiwei
Li, Zongchen
Wang, Yanlin
Chen, Chong
Ma, Yuchi
contents Training effective software engineering agents requires large volumes of task-specific trajectories, incurring substantial data construction costs. Inspired by the "Less-Is-More" hypothesis in mathematical reasoning, we investigate its extension to agentic scenarios and propose an end-to-end training framework that achieves superior agentic capabilities with fewer but higher-quality training trajectories. This is achieved via STITCH (Sliding-memory Trajectory Inference and Task Chunking Heuristic), a coarse-to-fine mechanism that filters low-value noise and retains decision-critical tokens to maximize training signal quality. We conduct experiments across multiple agent frameworks (e.g., mini-SWE-agent, MSWE-agent), model scales (30B to 355B), and multilingual settings (Python, Java, and ArkTS). On SWE-bench Verified, models trained with STITCH achieve up to 63.16% relative improvement over base models. On Multi-SWE-bench (Java), MiniMax-M2.5-STITCH achieves 43.75% with our CodeArts Agent scaffold (+16.67%). On HarmonyOS (ArkTS), GLM-4.7-STITCH improves the compilation pass rate to 61.31% (+43.34%) with less than 1K training trajectories. Our results confirm that the "Less-Is-More" paradigm generalizes effectively to complex agentic tasks across diverse languages and model scales.
format Preprint
id arxiv_https___arxiv_org_abs_2604_00824
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Yet Even Less Is Even Better For Agentic, Reasoning, and Coding LLMs
CodeArts Model Team
Ye, Yang
Tan, Jingyuan
Jiang, Tianyue
Ye, Ruizhe
He, Qiankun
Yang, Jiarui
Dong, Jian
Liang, Sicong
Yue, Chongjian
Xu, Peibai
Lu, Lufan
Pang, Shiguan
Qian, Taotao
Hu, Junbao
Hao, Yuechan
Shi, Ensheng
Zhang, Qi
Hao, Yi
Fan, Na
Tan, Xin
Yao, Shuai
Shen, Zhiwei
Li, Zongchen
Wang, Yanlin
Chen, Chong
Ma, Yuchi
Software Engineering
Training effective software engineering agents requires large volumes of task-specific trajectories, incurring substantial data construction costs. Inspired by the "Less-Is-More" hypothesis in mathematical reasoning, we investigate its extension to agentic scenarios and propose an end-to-end training framework that achieves superior agentic capabilities with fewer but higher-quality training trajectories. This is achieved via STITCH (Sliding-memory Trajectory Inference and Task Chunking Heuristic), a coarse-to-fine mechanism that filters low-value noise and retains decision-critical tokens to maximize training signal quality. We conduct experiments across multiple agent frameworks (e.g., mini-SWE-agent, MSWE-agent), model scales (30B to 355B), and multilingual settings (Python, Java, and ArkTS). On SWE-bench Verified, models trained with STITCH achieve up to 63.16% relative improvement over base models. On Multi-SWE-bench (Java), MiniMax-M2.5-STITCH achieves 43.75% with our CodeArts Agent scaffold (+16.67%). On HarmonyOS (ArkTS), GLM-4.7-STITCH improves the compilation pass rate to 61.31% (+43.34%) with less than 1K training trajectories. Our results confirm that the "Less-Is-More" paradigm generalizes effectively to complex agentic tasks across diverse languages and model scales.
title Yet Even Less Is Even Better For Agentic, Reasoning, and Coding LLMs
topic Software Engineering
url https://arxiv.org/abs/2604.00824