KnowU-Bench - Proactive (Easy): leaderboard
Proactive mobile assistance on KnowU-Bench (2026), easy split: 32 situations with no instruction, where the agent must decide from the user logs and current context whether to act, ask the simulated user for consent or stay silent, and must respect a rejection. Success rate in percent; higher is better.
Metric: Success rate (%) on the 32 easy proactive tasks (no instruction: act, ask for consent or stay silent, and respect a rejection) in the reproducible KnowU-Bench Android emulator, noisy full-history memory setting (the complete user logs plus irrelevant history), gpt-4o as the user simulator grounded in a hidden user profile, scored by rule-based checks fused with a rubric-conditioned LLM judge; higher is better. Source: arxiv.org. Saturation forecast: Around June 2027. 11 models tracked. Updated 7 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 84.4 | #173 |
| 2 | Qwen 3.5 397B A17B | 68.8 | #264 |
| 3 | Qwen 3.5 9B | 65.6 | #828 |
| 4 | Seed 2.0 Pro | 62.5 | #222 |
| 5 | Gemini 3.1 Pro (Preview) | 50 | #99 |
| 6 | Qwen 3 VL 32B | 50 | #622 |
| 7 | Qwen 3 VL 8B | 46.9 | #917 |
| 8 | UI-Venus-1.5-8B | 34.4 | |
| 9 | GUI-OWL-1.5-8B | 28.1 | |
| 10 | Qwen 3.5 122B A10B | 25 | #346 |
| 11 | MAI-UI-8B | 17.9 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=knowu-bench-proactive-easy · How It Works · Data refreshed daily, snapshot 2026-11-21.