OSWorld

OSWorld is a benchmark that contains configurations for multi-application environments used to evaluate the performance of computer-use agents (CUAs) in interacting with graphical desktops.

Papers using OSWorld (28)

Learn from Weaknesses: Automated Domain Specialization for Small Computer-Use Agents2026Mobile-agent-v3.5: Multi-platform Fundamental GUI Agents2026OS-Symphony: A Holistic Framework for Robust and Generalist Computer-Using Agent2026Multi-Agent Computer Use2026MacArena: Benchmarking Computer Use Agents on an Online macOS Environment2026Windows Agent Arena: Evaluating Multi-modal OS Agents At Scale2024 · 2 citesiOSWorld: A Benchmark for Personally Intelligent Phone Agents2026OSGuard: A Benchmark for Safety in Computer-Use Agents2026Infantagent-next: A Multimodal Generalist Agent For Automated Computer Interaction2025ProCUA-SFT Technical Report2026OSWorld-Human: Benchmarking the Efficiency of Computer-Use Agents2026Infiniteweb: Scalable Web Environment Synthesis For GUI Agent Training2026IntentScore: Intent-Conditioned Action Evaluation for Computer-Use Agents2026Agent Alpha: Tree Search Unifying Generation, Exploration and Evaluation for Computer-Use Agents2026Agent Skills for Large Language Models: Architecture, Acquisition, Security, and the Path Forward2026CaMeLs Can Use Computers Too: System-level Security for Computer Use Agents2026BEAP-Agent: Backtrackable Execution and Adaptive Planning for GUI Agents2026Grounding Computer Use Agents on Human Demonstrations2025Agentic Lybic: Multi-Agent Execution System with Tiered Reasoning and Orchestration2025CoAct-1: Computer-using Multi-Agent System with Coding Actions2025Surfer 2: The Next Generation Of Cross-platform Computer Use Agents2025Mano Technical Report2025Scaling Agents For Computer Use2025Instruction Agent: Enhancing Agent With Expert Demonstration2025UI-TARS-2 Technical Report: Advancing GUI Agent With Multi-turn Reinforcement Learning2025Ui-evol: Automatic Knowledge Evolving For Computer Use Agents2025Osworld: Benchmarking Multimodal Agents For Open-ended Tasks In Real Computer Environments2024 · 9 citesAgent S: An Open Agentic Framework that Uses Computers Like a Human2024 · 2 cites