# OSWorld-2.1 (Offline subset) Contamination Audit

> OSWorld-2.1 (Offline subset) scores 5.2/10 (Tier B, Moderate Risk) on BenchAudit as of November 2026, reflecting auto-approved click-through or encrypted public access and self-reported or live-network evaluation.

- **Canonical URL:** https://benchaudit.org/benchmarks/osworld-2-1-offline/
- **Benchmark ID:** `osworld-2-1-offline`
- **Hub:** OSWorld / XLANG Lab (HKU)
- **Category:** Knowledge Work
- **Domains:** Business, Coding, Finance, Science
- **Skills:** Agentic, Tool Use, Multimodal, Coding, Knowledge Work, Reasoning
- **BenchAudit Score:** 5.2/10 (Tier B - Moderate Risk)
- **Last Audited:** 2026-11-19 (Ref 2026-11)

## Factor Scores

| Factor | Score (/10) | Weight |
| --- | ---: | ---: |
| Privacy | 2 | 40% |
| Originality | 10 | 25% |
| Freshness | 7 | 20% |
| Grading | 3.3 | 15% |

## Primary-Source Evidence

### Privacy
> "Visit https://huggingface.co/datasets/xlangai/osworld_v2_tasks while logged in to Hugging Face and accept the gated access request, your request will automatically be approved."
- Source: https://github.com/xlang-ai/OSWorld-V2

### Originality
> "The bulk of OSWorld 2.0 tasks were produced by a small group of internally trained annotators who worked end-to-end on the tasks they proposed, covering task design, input artifact construction, environment setup, and evaluation function implementation."
- Source: https://arxiv.org/html/2606.29537v2

### Freshness
> "2026-09-16: We released osworld-v2.1, a new bug-fix release with updated agents, evaluation workflows, and pinned task, asset, website, and provider image references."
- Source: https://github.com/xlang-ai/OSWorld-V2

### Grading
> "Offline set: 82 tasks runnable without internet"
- Source: https://osworld-v2.xlang.ai/static/js/leaderboard.js
