AndroidIntent - Proactive Suggestions: leaderboard
Proactive-service setting of AndroidIntent (PersonalAlign, Lyu et al., Harbin Institute of Technology, Shenzhen, ACL 2026): from a user's recent phone records and current state, the agent proposes the action the user is likely to want next, on 215 routine interactions. A DeepSeek-V3 judge decides whether each suggestion matches the user's actual intent; percent, higher is better.
Metric: Intent alignment judged by DeepSeek-V3 (%; proactive suggestions from recent user records and the current user state, 215 routine interactions). Source: arxiv.org. Saturation forecast: Around 2031. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 37 | #135 |
| 2 | Qwen Max | 35.4 | #990 |
| 3 | Qwen-VL-Max | 34.8 | #1226 |
| 4 | GLM-4.5V | 33.9 | #739 |
| 5 | GPT-5.1 | 32 | #225 |
| 6 | Qwen 3 8B | 26.8 | #1838 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=androidintent-proactive-suggestions · How It Works · Data refreshed daily, snapshot 2026-11-12.