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.
@article{arxiv.2604.00824,
title = {Yet Even Less Is Even Better For Agentic, Reasoning, and Coding LLMs},
author = {CodeArts Model Team and Yang Ye and Jingyuan Tan and Tianyue Jiang and Ruizhe Ye and Qiankun He and Jiarui Yang and Jian Dong and Sicong Liang and Chongjian Yue and Peibai Xu and Lufan Lu and Shiguan Pang and Taotao Qian and Junbao Hu and Yuechan Hao and Ensheng Shi and Qi Zhang and Yi Hao and Na Fan and Xin Tan and Shuai Yao and Zhiwei Shen and Zongchen Li and Yanlin Wang and Chong Chen and Yuchi Ma},
journal= {arXiv preprint arXiv:2604.00824},
year = {2026}
}