GMA - Compositional: leaderboard
Metric: Success rate (%; 300 tasks in seven custom Android applications, vision-only agent seeing raw screenshots (two most recent kept), MobileWorld action space plus call_user and answer, Qwen3.7-Plus user simulator, 150-step budget; 129 compositional multi-step tasks in one application). Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.8 Max | 70.54 |
| 2 | GPT-5.6 Sol | 68.99 |
| 3 | Seed 2.1 Pro | 65.89 |
| 4 | GPT-5.5 | 65.12 |
| 5 | Claude Opus 4.7 | 65.12 |
| 6 | Gemini 3.5 Flash | 62.02 |
| 7 | Claude Sonnet 4.6 | 60.47 |
| 8 | Qwen 3.7 Plus | 51.94 |
Interactive version: theaggregate.ai/benchmark?slug=gma-compositional · How It Works · Data refreshed daily, snapshot 2026-09-29.