DevicesWorld - Mean Score: leaderboard
DevicesWorld (2026) is an executable benchmark of cross-device tasks in which an agent gathers information and acts across Android, Linux desktop and smart-home devices within 50 steps, verified from final device states and files. Mean share of the scoring conditions satisfied per task, in percent; higher is better.
Metric: Mean partial score (%; per task, the share of enabled scoring conditions satisfied, averaged over tasks; fixed stratified evaluation set drawn from 6,140 executable cross-device tasks spanning Android, Linux desktop and SmartHome devices, with rule, state and file verifiers; at most 50 interaction steps, temperature 0). Source: arxiv.org. Saturation forecast: Around February 2028. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 26.2 | #49 |
| 2 | Gemini 3.1 Pro (Preview) | 25.2 | #94 |
| 3 | Claude Opus 4.8 | 21.3 | #56 |
| 4 | Qwen 3.7 Plus | 20.1 | #207 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=devicesworld-mean-score · How It Works · Data refreshed daily, snapshot 2026-11-13.