MobileJudgeBench

931 human-annotated mobile-agent trajectories for evaluating LLM-as-judge methods.

Paper: Benchmarking LLM Judges for Mobile Agent Evaluation (arXiv:2608.11434) · Code and loader: github.com/WZq975/MobileJudgeBench

MobileJudgeBench is the benchmark introduced in Benchmarking LLM Judges for Mobile Agent Evaluation. Mobile-agent benchmarks and training increasingly rely on LLM judges to decide whether an agent completed a task; this dataset lets you measure how reliable such judges are. Every trajectory is a full record of one agent attempting one task on an Android emulator (screenshots, accessibility trees, actions, the agent's reasoning) together with a human success/failure label.

Trajectories	931 (492 success / 439 failure)
Tasks	289 tasks over 59 apps, from SPA-Bench, AndroidWorld, A3, AndroidArena, B-MoCA and AndroidLab
Agents	GPT-5-mini and Qwen2.5-VL-72B (M3A), Llama-3.3-70B (T3A), UI-TARS-72B
Steps	11,502 agent steps, 11,911 screenshots (1080x2400 PNG)
Labels	binary task-success label from human annotation (9 trained annotators, 2-4 per trajectory)
License	data CC BY 4.0, code MIT

Download

The trajectory data is 4.6 GiB (PNG screenshots), distributed as 24 tar archives (one per benchmark x agent, each under 1 GiB) listed in manifest.json with SHA-256 sums; they are hosted on the Hugging Face Hub at huggingface.co/datasets/WesleyWzq/MobileJudgeBench. The index files (annotations.csv, tasks.csv) sit next to the archives and are shown in the dataset viewer. The code repository (github.com/WZq975/MobileJudgeBench) holds the downloader, the loader and the paper's Baseline judge:

git clone https://github.com/WZq975/MobileJudgeBench.git
cd MobileJudgeBench

# everything (~4.6 GiB)
python -m mobilejudgebench.download --dest ./MobileJudgeBench

# subsets
python -m mobilejudgebench.download --dest ./MJB --benchmark spa_bench --benchmark a3
python -m mobilejudgebench.download --dest ./MJB --agent ui-tars-72b
python -m mobilejudgebench.download --dest ./MJB --metadata_only

After downloading, python scripts/verify_release.py --root ./MobileJudgeBench --expect_trajectories 931 checks that everything is complete and consistent.

Quick start

from mobilejudgebench import load_annotations, load_trajectory, iter_trajectories

root = "./MobileJudgeBench"
rows = load_annotations(root)                       # list of dicts, one per trajectory
traj = load_trajectory(root, rows[0]["trajectory_id"])
print(traj.goal, traj.success, traj.n_steps)
for step in traj.steps:
    print(step.index, step.action, step.screenshot_path)
    # step.reasoning, step.ui_elements (None for UI-TARS)
images = traj.load_images()                         # PIL images in time order (the judge input)

for traj in iter_trajectories(root, benchmark="android_world", agent="gpt-5-mini"):
    ...

examples/quickstart.py --root ./MobileJudgeBench prints dataset statistics and renders one trajectory as a contact sheet. The loader has no third-party dependency (Pillow only for images).

To evaluate a judge, compare its binary predictions with the success column of annotations.csv (positive class = successful trajectory); the paper reports accuracy, precision, recall, F1 and balanced accuracy. The judge pipeline in the code repository (mobilejudgebench.judges, see its docs/JUDGES.md) runs the paper's Baseline judge or your own judge function over this data, and scores verdict files produced elsewhere.

Data format

trajectories/<benchmark>/<agent>/<task_id>/
├── trajectory.json      goal, human label, per-step action and reasoning
├── ui_elements.json     accessibility-tree elements for each screenshot (M3A and T3A agents)
└── screenshots/
    ├── step_000.png     screen before step 0 (one per step)
    ├── ...
    └── step_<n>.png     final screen, when available

Benchmark ids: spa_bench, android_world, a3, android_arena, bmoca, android_lab. Agent ids: gpt-5-mini, qwen2.5-vl-72b, llama-3.3-70b, ui-tars-72b. annotations.csv holds one row per trajectory (success = human label, plus task, app and step-count metadata); tasks.csv lists the 289 tasks. Field-level documentation and provenance notes: FORMAT.md.

Files

MobileJudgeBench-v1.0.0-<benchmark>-<agent>.tar   trajectory directories (24 archives)
MobileJudgeBench-v1.0.0-metadata.tar.gz           this README, FORMAT.md, licenses, annotations.csv, tasks.csv
manifest.json                                     archive -> trajectory ids, sizes, SHA-256
SHA256SUMS                                        SHA-256 of every archive
CHECKSUMS_files.sha256                            SHA-256 of every file inside the extracted release
annotations.csv, tasks.csv, FORMAT.md             copies of the index files and documentation

License and attribution

The trajectories, annotations and documentation are released under CC BY 4.0; the code under the MIT License.

The tasks and app configurations originate from the source benchmarks, which we gratefully acknowledge: AndroidWorld (Apache-2.0), B-MoCA (Apache-2.0), SPA-Bench (MIT), A3 (Apache-2.0), AndroidLab (MIT) and AndroidArena (task definitions used for research evaluation with attribution). Agents: AndroidWorld M3A/T3A (Apache-2.0) and UI-TARS (Apache-2.0).

Citation

@misc{wang2026benchmarkingllmjudgesmobile,
  title         = {Benchmarking LLM Judges for Mobile Agent Evaluation},
  author        = {Ziqiang Wang and Li Gu and Zhixiang Chi and Zhi Liu and Seyed Mehdi Ayyoubzadeh and Yuanhao Yu and Yang Wang},
  year          = {2026},
  eprint        = {2608.11434},
  archivePrefix = {arXiv},
  primaryClass  = {cs.AI},
  url           = {https://arxiv.org/abs/2608.11434}
}

Changelog

• v1.0.0 - initial public release: 931 trajectories, 289 tasks, human success labels.