Research map / Multi-agent systems and games / LLM agents
Computer Agent Arena: Toward Human-Centric Evaluation and Analysis of Computer-Use Agents
Bowen Wang, Xinyuan Wang, Jiaqi Deng, Tianbao Xie, Ryan Li, Yanzhe Zhang, Junli Wang, Dunjie Lu, Zicheng Gong, Gavin Li, Toh Jing Hua, Wei-Lin Chiang and 7 more
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Presents Computer Agent Arena, an open-source platform for head-to-head, user-driven evaluation of computer-use agents, and a method that converts human judgments into dynamic evaluation.
Problem
Static, manually curated CUA benchmarks are narrow, contamination-prone, environment-heavy, and diverge from real user-driven evaluation.
Approach
Head-to-head arena-style comparison of computer-use agents with human judgments converted into a dynamic evaluation methodology, shifting CUA evaluation toward user-driven, human-centric assessment.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| evaluate on human-submitted open-domain tasks instead of static OSWorld tasks | CUA leaderboard · Computer Agent Arena | model ranking | Mixed |
| split tasks into OSWorld in-domain vs out-of-domain (ablation) | CUA Elo ranking · Computer Agent Arena (1,000 tasks) | Elo ranking | Mixed |
| moderate CALL_USER clarification queries (1-2) vs none or many (ablation) | CUA agents · Computer Agent Arena | win rate | Improved |
| add tool/code-integrated action space to CUA | CoAct-1 · Computer Agent Arena | user preference / Elo: OSWorld-Verified 60.1% vs Elo 1043 (rank 5) | Hurt |
| faster or shorter execution when both agents are correct (ablation) | CUA agents · Computer Agent Arena | user preference | No clear effect |
| higher task correctness (ablation) | CUA agents · Computer Agent Arena | win rate | Improved |
Closest papers by meaning
- Computer Use at the Edge of the Statistical Precipice (NeurIPS 2026 · similarity 0.92)
- OpenCUA: Open Foundations for Computer-Use Agents (NeurIPS 2025 · similarity 0.92)
- Windows Agent Arena: Evaluating Multi-Modal OS Agents at Scale (ICML 2025 · similarity 0.90)
- Just Do It!? Computer-Use Agents Exhibit Blind Goal-Directedness (ICLR 2026 · similarity 0.90)
- Agent S: An Open Agentic Framework that Uses Computers Like a Human (ICLR 2025 · similarity 0.89)
- Talk, Evaluate, Diagnose: User-aware Agent Evaluation with Automated Error Analysis (ICLR 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
