multimodal benchmark
OSWorld Screenshot-only
OSWorld Screenshot-only: A variant of the OSWorld benchmark that evaluates multimodal AI agents using only screenshot observations to complete open-ended computer tasks across real operating systems (Ubuntu, Windows, macOS). Tests agents' ability to perform complex workflows involving web apps, desktop applications, file I/O, and multi-application tasks through visual interface understanding and GUI grounding.
Updated Aug 7, 2026
Published models1
Registry coverage1
MetricScore
EvidenceB
OSWorld Screenshot-only leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | AN | 14.9% | 100.0% | 1 | C |
OSWorld Screenshot-only highlights
The top published results on this benchmark's own scale.
What is OSWorld Screenshot-only?
Definition and scoring fields from the benchmark registry.
OSWorld Screenshot-only: A variant of the OSWorld benchmark that evaluates multimodal AI agents using only screenshot observations to complete open-ended computer tasks across real operating systems (Ubuntu, Windows, macOS). Tests agents' ability to perform complex workflows involving web apps, desktop applications, file I/O, and multi-application tasks through visual interface understanding and GUI grounding.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- OSWorld Screenshot-only
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- osworld-screenshot-only|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about OSWorld Screenshot-only.
Which model scores highest on OSWorld Screenshot-only?
Claude 3.5 Sonnet is currently ranked first with 14.9%.
What does OSWorld Screenshot-only measure?
OSWorld Screenshot-only: A variant of the OSWorld benchmark that evaluates multimodal AI agents using only screenshot observations to complete open-ended computer tasks across real operating systems (Ubuntu, Windows, macOS). Tests agents' ability to perform complex workflows involving web apps, desktop applications, file I/O, and multi-application tasks through visual interface understanding and GUI grounding.
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.