MAI-UI-8B: benchmark results
Provider: Microsoft. Access: Open.
Unified ELO 1705 ± 21, rank #225 of 1605 rated models, from 20 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MobileWorldSafety - Attack Success Rate | 40.4 | Attack success rate (%; share of scorable tasks in which the | 100 |
| DocOS - Easy | 20.99 | Task completion rate (%; share of tasks whose final applicat | 80 |
| DocOS - Hard | 10 | Task completion rate (%; share of tasks whose final applicat | 80 |
| DocOS - Medium | 7.82 | Task completion rate (%; share of tasks whose final applicat | 80 |
| MobileWorldSafety - Task Completion Rate | 47.2 | Task completion rate under attack (%; share of scorable task | 80 |
| Memory-World - History-Retention Protocol | 40.2 | Task Accuracy (%; pass@1, GUI-specialized agents, native pro | 75 |
| CLI-Advantage Suite (GUI Agents) | 27.4 | Success rate (%; 45 task templates x 3 seeds = 135 instances | 50 |
| AndroidWorld | 70.7 | Success Rate pass@1 (%) | 48.8 |
| PhoneSafety | 29.4 | Safe-action rate (%; 700 safety-critical moments; neither sa | 42.9 |
| PhoneSafety - Trap Resistance | 85.9 | Safe-action rate (%; 78 trap-resistance moments) | 42.9 |
| UI-Venus-2 Report - MemGUI | 17.2 | Pass@1 (%) | 42.9 |
| MobileWorld - MCP | 20 | Success Rate (%) | 37.5 |
Interactive version: theaggregate.ai/model?slug=mai-ui-8b · How It Works · Data refreshed daily, snapshot 2026-09-26.