multimodal benchmark
OSWorld 2.0
OSWorld 2.0 is a benchmark of 108 long-horizon, real-world computer-use workflows spanning everyday and professional tasks. Each task is an end-to-end workflow that takes human users a median of about 1.6 hours, scored with a binary-completion metric, and targets challenges such as dynamic environments, cross-source reasoning, and implicit-state inference.
Updated Aug 11, 2026
Models4
Model coverage4
MetricScore
EvidenceB
OSWorld 2.0 Ranking
Higher score ranks better on this benchmark.
| 1 | AN | 70.6% | 100.0% | 4 | C | |
| 2 | OP | 62.6% | 66.7% | 4 | C | |
| 3 | OP | 50.2% | 33.3% | 4 | C | |
| 4 | OP | 45.6% | 0.0% | 4 | C |
OSWorld 2.0 Score Distribution
A closer view of the leading scores on this benchmark.
OSWorld 2.0 Highlights
The leading models and scores on this benchmark.
What is OSWorld 2.0?
What OSWorld 2.0 measures and how its scores work.
OSWorld 2.0 is a benchmark of 108 long-horizon, real-world computer-use workflows spanning everyday and professional tasks. Each task is an end-to-end workflow that takes human users a median of about 1.6 hours, scored with a binary-completion metric, and targets challenges such as dynamic environments, cross-source reasoning, and implicit-state inference.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.
- Family
- OSWorld 2.0
- Modality
- multimodal
- Primary category
- multimodal
- Score direction
- higher
- LLMBoard eligible
- Yes
- Evaluation key
- osworld-2.0|llm-stats-current
Benchmark scores retain their original unit. Overall score eligibility is shown separately.
FAQ
Common questions about OSWorld 2.0.
Which model scores highest on OSWorld 2.0?
Claude Opus 5 is currently ranked first with 70.6%.
What does OSWorld 2.0 measure?
OSWorld 2.0 is a benchmark of 108 long-horizon, real-world computer-use workflows spanning everyday and professional tasks. Each task is an end-to-end workflow that takes human users a median of about 1.6 hours, scored with a binary-completion metric, and targets challenges such as dynamic environments, cross-source reasoning, and implicit-state inference.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
4 model results are currently shown.
Does this benchmark affect the overall score?
Yes. This benchmark can contribute to the current LLMBoard capability score.