AppSim-Bench: leaderboard
Mobile GUI agents on reproducible simulations of 18 real Chinese and English apps (Bilibili, Ctrip, WeChat, Amazon, Booking, YouTube and others); task-weighted accuracy over the whole benchmark. Higher is better. Source: arxiv.org/abs/2609.07712.
Metric: Task-weighted Accuracy (%). Source: arxiv.org. Saturation forecast: Around March 2027. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 50.27 | #51 |
| 2 | Gemini 3 Pro | 49.91 | #82 |
| 3 | GPT-5.5 | 49.73 | #33 |
| 4 | Gemini 3.1 Pro (Preview) | 49.55 | #64 |
| 5 | Seed-1.6 | 39.14 | #272 |
| 6 | GPT-5 | 33.39 | #105 |
| 7 | Seed 1.8 | 32.14 | #143 |
| 8 | Qwen 3.6 Plus | 31.42 | #124 |
| 9 | GPT-5.4 | 29.08 | #83 |
| 10 | Qwen 3.6 Flash | 28.19 | #221 |
| 11 | Claude Sonnet 4.6 | 27.29 | #94 |
| 12 | Claude Haiku 4.5 | 22.8 | #287 |
| 13 | Claude Sonnet 4.5 | 21.01 | #148 |
| 14 | Gemini 2.5 Pro | 17.24 | #159 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=appsim-bench · How It Works · Data refreshed daily, snapshot 2026-11-05.