ContrAR (Single Frame): leaderboard

ContrAR (2026) tests whether vision-language models detect contradictory virtual content attacks in augmented reality, answering yes or no on real AR videos of which half carry an overlay that contradicts the physical scene, given the middle frame of each video. Detection accuracy in percent over five AR applications; higher is better.

Metric: Detection accuracy (%) for contradictory virtual content attacks in AR videos (all 312 videos of five AR applications; 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 February 2027. 11 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-588.14#178
2Gemini 2.5 Pro83.97#276
3GPT-4.182.05#505
4Gemini 2.5 Flash79.81#463
5GPT-4o79.17#751
6Grok 468.27#319
7Grok 2 Vision64.74#1437
8Claude Sonnet 4.563.78#261
9Qwen 2.5 VL 72B Instruct60.26#841
10Qwen 2.5 VL 7B Instruct58.01#1902
11Claude Haiku 4.553.85#583

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=contrar-single-frame · How It Works · Data refreshed daily, snapshot 2026-11-21.