MobileGym-Bench - Deep-Dive Tasks: leaderboard
Metric: Success rate (%; 68 of the 256 test tasks, state-checked; open models: mean of four runs with re-sampled task parameters; proprietary models: one run (two for Gemini 3.1 Pro)). Source: arxiv.org. Saturation forecast: Around December 2026. 9 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 43.4 |
| 2 | Seed 2.0 Pro | 36.8 |
| 3 | Qwen 3.6 Plus | 32.4 |
| 4 | uitars-1.5-7B | 4.8 |
| 5 | Qwen 3 VL 4B Instruct | 1.5 |
Interactive version: theaggregate.ai/benchmark?slug=mobilegym-bench-deep-dive-tasks · How It Works · Data refreshed daily, snapshot 2026-09-25.