OSWorld-Verified

OSWorld-Verified measures computer use: the model controls a real desktop (mouse, keyboard, GUI) inside a virtual OS and has to complete everyday tasks like filling forms, managing files or configuring software. The task set went through a verification pass so that every task has clearly checkable success criteria.

OSWorld-Verified measures how well a model acts as a general-purpose agent. In detail, it checks:

• seeing and understanding what is on screen
• clicking, typing and scrolling through a real GUI
• composing single clicks into a multi-step solution
• reaching a verifiable final state (e.g. the saved file looks as required)

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better.

What is the OSWorld-Verified benchmark?

OSWorld-Verified measures computer use: the model controls a real desktop (mouse, keyboard, GUI) inside a virtual OS and has to complete everyday tasks like filling forms, managing files or configuring software. The task set went through a verification pass so that every task has clearly checkable success criteria.

What does OSWorld-Verified test?

OSWorld-Verified mainly measures how well a model acts as a general-purpose agent. Concretely, it checks: seeing and understanding what is on screen; clicking, typing and scrolling through a real GUI; composing single clicks into a multi-step solution.

How is the OSWorld-Verified score calculated?

The score is accuracy: the percentage of tasks answered or solved correctly. 100% means every task was solved perfectly; higher is better. The declared maximum score of this benchmark is 100.0.

Where do I get OSWorld-Verified data and scores?

The official benchmark page/reference is linked on this page. Every score on Open LLM Radar is also available as open data: download the scores JSON (http://open-llm-radar-api:8000/api/v1/benchmarks/osworld-verified/scores) or pull it directly from the Open LLM Radar API with any HTTP client.