UI-Venus-2 Report - VenusBench-CAPTCHA: leaderboard
VenusBench-CAPTCHA has 219 screenshots across eight CAPTCHA interaction types, from OCR text entry and ordered clicks to rotation, slider and drag puzzles, scored as executable actions under strict checks. From the UI-Venus-2 technical report, every model run single-turn on the same fixed set; Pass@1 in percent, higher is better.
Metric: Pass@1 (%). Source: arxiv.org. Saturation forecast: Around 2032. 7 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 27B | 53 | #182 |
| 2 | Seed 2.0 Pro | 47.9 | #122 |
| 3 | Kimi K2.6 | 39.7 | #113 |
| 4 | Qwen 3.5 9B | 28.3 | #391 |
| 5 | Claude Opus 4.6 | 16 | #67 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ui-venus-2-report-venusbench-captcha · How It Works · Data refreshed daily, snapshot 2026-11-05.