DevicesWorld: leaderboard

Metric: Task success rate (%; a task succeeds only when every required verification item passes; 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 2029. 5 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)12
2GPT-5.512
3Claude Opus 4.810.5
4Qwen 3.7 Plus9.5

Interactive version: theaggregate.ai/benchmark?slug=devicesworld · How It Works · Data refreshed daily, snapshot 2026-09-29.