ContrAR (Single Frame) - Safety Inspection: leaderboard
Metric: Detection accuracy (%) for contradictory virtual content attacks in AR videos (42 safety inspection videos; half the videos carry a contradiction, a balanced yes/no decision), one middle frame of each video given (single-frame prompting); mean of three runs; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 11 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-4.1 | 90.48 |
| 2 | Gemini 2.5 Pro | 83.33 |
| 3 | GPT-5 | 80.95 |
| 4 | GPT-4o | 71.43 |
| 5 | Claude Sonnet 4.5 | 71.43 |
| 6 | Gemini 2.5 Flash | 69.05 |
| 7 | Claude Haiku 4.5 | 64.29 |
| 8 | Grok 4 | 64.29 |
| 9 | Qwen 2.5 VL 7B Instruct | 61.9 |
| 10 | Qwen 2.5 VL 72B Instruct | 50 |
Interactive version: theaggregate.ai/benchmark?slug=contrar-single-frame-safety-inspection · How It Works · Data refreshed daily, snapshot 2026-10-07.