TTableAI Catalog

benchmark

OSWorld

A benchmark for multimodal agents that operate full computer environments through visual interaction.

Visit projectView source

What it is

OSWorld supplies real operating-system tasks, virtual machine environments, and execution-based evaluation for computer-use agents.

Related records

SWE-benchA benchmark for evaluating language models on real software engineering issues from open-source repositories.Terminal-BenchA benchmark for measuring AI agents on practical tasks completed in terminal environments.AgentBenchA benchmark suite designed to evaluate language models acting as agents across varied environments.WebArenaA realistic web environment for evaluating autonomous agents on reproducible tasks.