Harness Scaling for Generalizable Agentic Reinforcement Learning
Fei Bai, Bryan Dai and colleagues at IQuest Research with Beihang University and Renmin University (Wayne Xin Zhao, Ji-Rong Wen) propose Harness Scaling, which trains agentic RL policies across many harness configurations instead of one.
Ask this paper
Problem. RL through a single fixed harness specializes the model to that harness and transfers poorly to other tool, context and control configurations.
Factorized harness space. The authors factor harnesses into components, select a compact training subset with balanced factorial coverage, and pair global coverage with controlled variation on the same task.
Generalization. Across benchmarks, model scales and RL algorithms, training this way improves performance on unseen white-box harness configurations.
Transfer to black-box harnesses. Models trained on structured white-box variation transfer to unseen external black-box harnesses better than models trained through one complex black-box harness.
Abstract
Agent harnesses have become a central component of modern agentic systems, shaping how models use tools, manage context, and coordinate long-horizon interaction. Realizing the benefits of these harnesses depends on how effectively the underlying model can make use of them, motivating reinforcement learning directly through the harness. However, existing approaches typically train through a single fixed harness, which can lead to specialization and limited transfer across harness configurations. To address this, we introduce \textbf{Harness Scaling}, a new scaling dimension for agentic RL that extends training from a single harness to diverse execution configurations. Concretely, we construct a factorized harness space, select a compact training subset with balanced factorial coverage, and organize cross-harness training to combine global coverage with controlled variation for the same task. Across multiple benchmarks, model scales, and RL algorithms, Harness Scaling consistently improves generalization to unseen white-box configurations. Moreover, models trained through structured white-box variation transfer effectively to unseen external black-box harnesses, outperforming models trained through a single complex black-box harness. These results establish harness diversity as a meaningful scaling dimension for agentic RL and highlight the importance of how this diversity is organized for cross-harness generalization.