Evo-Bench: Can Language Models Improve Agent Harness?
2026-08-10 • Computation and Language
Computation and Language
AI summaryⓘ
The authors created Evo-Bench, a new test to measure how well language models can improve their own tools and strategies (called harnesses) without relying on better base models or overfitting to specific tasks. They designed it to fairly test models across different areas like Search, Office, and general tasks by carefully selecting sensitive tasks and ensuring results generalize well. Their tests showed that while models do very well with general and search tasks, they have trouble with office-like tasks that need specific workflows. They also found that the improved harnesses help models think better across different tasks and that there are issues like improvements hitting a limit early on.
Large Language Modelsautonomous agentsharness evolutionbenchmarkingtask overfittingcross-suite generalizationauxiliary-task evolutiontemporal anomaliestransferable reasoningpolicy models
Authors
Lisheng Huang, Chen Yang, Hao Zhou, Huatong Song, Zongchao Chen, Ran Le, Yang Song, Wayne Xin Zhao, Tao Zhang
Abstract
Large Language Models (LLMs) have driven rapid progress in autonomous agents, yet standard evaluations remain confined to static task solving. An emerging frontier is harness evolution---the agent's capacity to autonomously optimize its own operating harness. However, systematically benchmarking this capability remains challenging, as existing evaluations fail to isolate harness improvements from base model strength, prevent task-specific overfitting, or capture long-horizon iterative research. To address these challenges, we introduce Evo-Bench, the first benchmark designed to evaluate models' intrinsic harness-evolving capabilities across Search, Office, and General agent domains. To rigorously isolate this capability, Evo-Bench employs a novel harness-guided construction framework: it leverages auxiliary-task evolution to identify tasks genuinely sensitive to framework improvements, followed by sensitivity-aware stratified splitting to ensure robust cross-suite generalization. Extensive evaluations across nine frontier and open-weight models reveal that top models achieve massive absolute gains reaching 16.6 points, closely approaching state-of-the-art human-engineered baselines. Crucially, while autonomous evolution outpeforms artificial harness in General tasks and excels in Search tasks, it struggles in Office tasks that demand highly specific processing workflows. Furthermore, our analysis exposes critical temporal anomalies like early saturation, while demonstrating that the synthesized harnesses act as highly transferable reasoning structures, consistently boosting diverse policy models.