Yet Even Less Is Even Better For Agentic, Reasoning, and Coding LLMs

TL;DR

This paper introduces STITCH, a training framework that enhances agentic reasoning and coding capabilities in large language models using fewer high-quality trajectories, inspired by the 'Less-Is-More' hypothesis.

Contribution

The authors extend the 'Less-Is-More' concept to agentic tasks and develop STITCH, a coarse-to-fine filtering mechanism that improves training efficiency and performance across multiple models and languages.

Findings

Models trained with STITCH show up to 63.16% improvement on SWE-bench.

MiniMax-M2.5-STITCH achieves 43.75% on Multi-SWE-bench Java.

GLM-4.7-STITCH increases compilation success rate to 61.31% on HarmonyOS.

Abstract

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.