OS-Themis: A Scalable Critic Framework for Generalist GUI Rewards

TL;DR

OS-Themis is a scalable critic framework for GUI reinforcement learning that improves reward accuracy and robustness by decomposing trajectories and auditing evidence, leading to better agent training outcomes.

Contribution

We introduce OS-Themis, a novel multi-agent critic framework that enhances reward evaluation for GUI agents and provide a new benchmark for cross-platform GUI reward assessment.

Findings

OS-Themis improves RL training performance by 10.3%.

OS-Themis enhances trajectory validation, yielding a 6.9% gain.

All evaluated models perform best under OS-Themis in the benchmark.

Abstract

Reinforcement Learning (RL) has the potential to improve the robustness of GUI agents in stochastic environments, yet training is highly sensitive to the quality of the reward function. Existing reward approaches struggle to achieve both scalability and performance. To address this, we propose OS-Themis, a scalable and accurate multi-agent critic framework. Unlike a single judge, OS-Themis decomposes trajectories into verifiable milestones to isolate critical evidence for decision making and employs a review mechanism to strictly audit the evidence chain before making the final verdict. To facilitate evaluation, we further introduce OmniGUIRewardBench (OGRBench), a holistic cross-platform benchmark for GUI outcome rewards, where all evaluated models achieve their best performance under OS-Themis. Extensive experiments on AndroidWorld show that OS-Themis yields a 10.3% improvement when used to support online RL training, and a 6.9% gain when used for trajectory validation and filtering in the self-training loop, highlighting its potential to drive agent evolution.