More Works from Our Team

VenusBench-GD Leaderboard

Multi-platform GUI grounding benchmark for diverse grounding tasks.

Leaderboard
UI-Venus Technical Report

High-performance UI agents with reinforcement fine-tuning.

arXiv 2025

Venus Team, Ant Group

VenusBench-Mobile Leaderboard

A challenging and user-centric benchmark for mobile GUI agents with capability diagnostics.

GitHub Paper Data
Primary Tasks 149
Stability Variants 80
Apps 27
Top SR --

Primary Success Rate

Success Rate (%) over the 149-task primary pool.

Loading data

Notes

  • Primary success rate values are reported as percentages on VenusBench-Mobile's 149-task primary pool.
  • SPR is the Stability Pass Rate: a base task passes only when all corresponding environment settings pass.
  • TT is total output tokens across tasks in thousands. PT is the average per-step output tokens.
  • Planner-executor rows use UI-Venus-72B as the grounding executor.
  • Rows added after the original manuscript may report only primary task success rate; unavailable fields are marked as N/A.

Citation

@misc{gong2026venusbenchmobile,
  title={VenusBench-Mobile: A Challenging and User-Centric Benchmark for Mobile GUI Agents with Capability Diagnostics},
  author={Yichen Gong and Zhuohan Cai and Sunhao Dai and Yuqi Zhou and Zhangxuan Gu and Changhua Meng and Shuheng Shen},
  year={2026},
  url={https://github.com/inclusionAI/UI-Venus/tree/VenusBench-Mobile}
}

Last updated on: May 11, 2026