MobileGym-Bench: leaderboard
Mobile-GUI agent benchmark over 256 Android tasks: success is graded by reading the simulator’s own JSON state after the run rather than by asking a second model to judge screenshots, so a task passes only when the device actually reaches the target state; success rate %.
Metric: Success Rate (%). Source: github.com. Saturation forecast: Around January 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 58.8 | #58 |
| 2 | Seed 2.0 Pro | 52 | #118 |
| 3 | Qwen 3.6 Plus | 45.7 | #120 |
| 4 | uitars-1.5-7B | 13.8 | #864 |
| 5 | Qwen 3 VL 4B Instruct | 9.4 | #515 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mobilegym-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.