# OSWorld

> OSWorld is a benchmark of open-ended tasks that computer-use agents must complete in real operating systems and applications, graded by scripts that check the resulting state of the computer.

- Identifier: PTL-0205
- Category: Benchmarks
- Canonical URL: https://protologue.com/t/osworld/
- Introduced: 2024

## Description

Tasks span office software, web browsers, file management, and multi-application workflows. It is the standard benchmark for computer-use agents and showed a large gap between early agents and people.

## Broader terms

- [Benchmark](https://protologue.com/t/benchmark/)

## Related terms

- [Computer-Use Agent](https://protologue.com/t/computer-use-agent/)
- [Multimodal Prompting](https://protologue.com/t/multimodal-prompting/)

## Sources

- Xie et al. (2024). OSWorld: Benchmarking Multimodal Agents for Open-Ended Tasks in Real Computer Environments. https://arxiv.org/abs/2404.07972

## Cite this entry

Protologue. (2026). OSWorld. In Protologue: A Taxonomy of Prompting and LLM Techniques (v1.4.0, PTL-0205). https://protologue.com/t/osworld/

License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)