Share your thoughts, 1 month free Claude Pro on usSee more
WorkDL logo mark

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

About

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.

Zehao Li, Zhenyu Wu, Yibo Zhao, Bowen Yang, Jingjing Xie, Zhaoyang Liu, Zhoumianze Liu, Kaiming Jin, Jianze Liang, Zonglin Li, Feng Wu, Bowen Zhou, Zun Wang, Zichen Ding• 2026

Related benchmarks

TaskDatasetResultRank
Reward ModelingOGRBench
Ubuntu Accuracy88.1
24
Trajectory completion judgmentOGRBench Windows
Accuracy85.5
20
Trajectory completion judgmentOGRBench MacOS
Accuracy93.5
20
Trajectory completion judgmentOGRBench Web
Accuracy87.4
20
Trajectory completion judgmentOGRBench Mobile
Accuracy89.9
20
Trajectory completion judgmentOGRBench Ubuntu
Accuracy86.8
20
Trajectory completion judgmentOGRBench (Overall)
Accuracy86.6
20
Reward ModelingAgentRewardBench
Precision76.8
12
Online Reinforcement LearningAndroidWorld
Success Rate60.34
9
Showing 9 of 9 rows

Other info

Follow for update