UltraVR - Anomaly Detection: leaderboard
The anomaly detection domain of UltraVR (2026): a composed grid of 100 industrial product images (MVTec LOCO) hides one cell that violates a stated normality rule, and the model must give its row and column. Models answer directly from the original ultra-resolution image under their own preprocessing; accuracy in percent, higher is better.
Metric: Accuracy (%) on the anomaly detection domain: locate the one logically or structurally anomalous cell (row and column) in a composed grid of 100 MVTec LOCO cells, direct answering (S0) with the original ultra-resolution image under each model default preprocessing (V2); four-choice questions except anomaly detection, which asks for the grid coordinate of the one anomalous cell out of 100; higher is better. Source: arxiv.org. Saturation forecast: Around April 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 30.8 | #49 |
| 2 | Gemini 3.1 Pro (Preview) | 30 | #94 |
| 3 | GPT-5.4 | 20.8 | #123 |
| 4 | Qwen 3.5 27B | 20.8 | #359 |
| 5 | Gemma 4 26B | 9.2 | #612 |
| 6 | GLM-4.5V | 5.7 | #744 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=ultravr-anomaly-detection · How It Works · Data refreshed daily, snapshot 2026-11-13.