MobileJudgeBench (SPA-Bench Judge): leaderboard
MobileJudgeBench (2026) measures how accurately an LLM judge decides from a recorded trajectory whether a mobile GUI agent completed its task, against human labels on 931 trajectories from six benchmarks. Accuracy in percent, higher is better, using the SPA-Bench judging method.
Metric: Judge accuracy (%; share of 931 human-annotated mobile-agent trajectories, from 6 benchmarks, 4 agent models and 68 apps, whose success or failure the judge classifies correctly; the SPA-Bench judge adapted to mobile trajectories, greedy decoding). Source: arxiv.org. Saturation forecast: Estimated already saturated. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini | 90.6 | #349 |
| 2 | Gemini 3 Flash (Preview) | 89.8 | #141 |
| 3 | Claude Sonnet 4.5 | 84.7 | #244 |
| 4 | GLM-4.6V | 82.2 | #698 |
| 5 | Qwen 2.5 VL 72B Instruct | 80.1 | #845 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mobilejudgebench-spa-bench-judge · How It Works · Data refreshed daily, snapshot 2026-11-12.