AIAny
Icon for item

OSReward: Instituting Standardized Evaluation for Cross-Platform Computer-Use Reward Models

Evaluates vision-language model judges on computer-using agent (CUA) trajectories to measure verifier reliability. Provides OSReward-Hard and OSReward-Multi challenge sets, the OS-Shepherd-100K reasoning-annotated corpus, and trained OS-Shepherd reward models that match commercial judges at ~30–60× lower cost.

Introduction

Why this matters CUAs record actions, states, and reasoning as they operate across GUIs and web platforms; deciding whether a trajectory actually fulfills an instruction is core to evaluation, dataset curation, and RL training. Existing manual verification is too slow to scale, and off-the-shelf vision-language judges show consistent failure modes. The paper's central insight is that judge reliability—not just raw model capability—drives whether automated verification can replace costly human annotation at training scale.

Key Findings
  • Systematic leniency bias: many state-of-the-art VLM judges frequently label failed runs as successes, creating a high false-success rate that distorts downstream evaluation and reward signals.
  • Benchmark design: OSReward bundles a realistic, human-verified corpus of CUA trajectories, plus two focused suites—OSReward-Hard for genuinely challenging cases and OSReward-Multi for fine-grained efficiency/alignment scoring—so evaluations stress practical failure modes.
  • Data + models: OS-Shepherd-100K is a reasoning-annotated training corpus; OS-Shepherd models (9B and 35B) are trained on it and provide low-cost, stable reward judgments that match commercial judges at roughly 30–60× lower inference cost on the presented frontier.
  • Analysis and robustness: the study includes input ablations, ensembling experiments, and held-out transfer tests that reveal when judges fail and when de-biasing generalizes to unseen benchmarks.
What this means (so what?)
  • For teams building CUA evaluation pipelines, relying on off-the-shelf VLM judges risks introducing systematic mislabels that can corrupt dataset labels and RL reward targets.
  • The OSReward benchmark lets practitioners measure these failure modes explicitly and compare judges under standardized, cross-platform conditions rather than toy or synthetic tasks.
  • OS-Shepherd provides a practical midground: open reward models that are affordable enough to run at training scale while achieving judge-level accuracy close to commercial systems, lowering the cost barrier for academic and industrial experimentation.
Suitable for and tradeoffs

Great fit if you need a realistic stress-test for automated trajectory judges, want a labeled corpus for training reward/judge models, or aim to replace costly human verification in RL pipelines. Look elsewhere if you require end-to-end production deployment guidance (this is a benchmark and model-release paper, not a turnkey deployment stack) or if your domain diverges strongly from GUI/web CUA interactions (results may not transfer to unrelated environments).

Information

  • Websitearxiv.org
  • OrganizationsThe University of Hong Kong, Xi’an Jiaotong University, Nanjing University, University of Science and Technology of China, National University of Singapore, Fudan University
  • AuthorsQiushi Sun, Kanzhi Cheng, Yian Wang, Bowen Yang, Hang Yan, Liheng Chen, Fangzhi Xu, Zichen Ding, Nuo Chen, Jialin Cao
  • Published date2026/07/30

More Items

Generates retrieval-centric Chain-of-Thought (RC-CoT) over initially retrieved candidates to improve unified multimodal retrieval via reranking or full-corpus re-retrieval with a dual-mode embedder. Trains an embedder–adviser framework (UniME-R1) using mined hard negatives, supervised learning, and retrieval-oriented reinforcement learning.

Replaces external environment interaction in agentic RL training with 'world rehearsal': the policy alternates between making tool calls and simulating their environment responses, jointly optimizing both roles so the agent internalizes environment dynamics and improves long-horizon tool use and transfer.

Generates large-scale, explorable 3D open-world scenes from open-ended text prompts, producing editable instance-level assets and a consistent global terrain. Uses agentic planning to convert text into region/terrain/asset specifications and a coarse-to-fine pipeline for terrain construction, mesh reconstruction, and render-based refinement.