multimodal benchmark
OSWorld Extended
OSWorld is a scalable, real computer environment benchmark for evaluating multimodal agents on open-ended tasks across Ubuntu, Windows, and macOS. It comprises 369 computer tasks involving real web and desktop applications, OS file I/O, and multi-application workflows. The benchmark evaluates agents' ability to interact with computer interfaces using screenshots and actions in realistic computing environments.
Updated Aug 7, 2026
Published models1
Registry coverage1
MetricScore
EvidenceB
OSWorld Extended leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AN | 22.0% | 100.0% | 1 | C |
OSWorld Extended highlights
The top published results on this benchmark's own scale.
What is OSWorld Extended?
Definition and scoring fields from the benchmark registry.
OSWorld is a scalable, real computer environment benchmark for evaluating multimodal agents on open-ended tasks across Ubuntu, Windows, and macOS. It comprises 369 computer tasks involving real web and desktop applications, OS file I/O, and multi-application workflows. The benchmark evaluates agents' ability to interact with computer interfaces using screenshots and actions in realistic computing environments.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- OSWorld Extended
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- osworld-extended|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about OSWorld Extended.
Which model scores highest on OSWorld Extended?
Claude 3.5 Sonnet is currently ranked first with 22.0%.
What does OSWorld Extended measure?
OSWorld is a scalable, real computer environment benchmark for evaluating multimodal agents on open-ended tasks across Ubuntu, Windows, and macOS. It comprises 369 computer tasks involving real web and desktop applications, OS file I/O, and multi-application workflows. The benchmark evaluates agents' ability to interact with computer interfaces using screenshots and actions in realistic computing environments.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
1 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.