MLVU-Test - Anomaly Recognition: leaderboard
MLVU-Test Anomaly Recognition (AR): the model must identify the unusual event in a long surveillance-style video, choosing among six options. Scored as accuracy in percent on the held-out long-video test set; higher is better (BAAI, CVPR 2025).
Metric: Multiple-choice accuracy (%). Source: github.com. Saturation forecast: Estimated already saturated. 35 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 68.8 | #741 |
| 2 | Qwen-VL-Max | 53.8 | #1226 |
| 3 | Claude 3 Opus | 30.8 | #1459 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mlvu-test-anomaly-recognition · How It Works · Data refreshed daily, snapshot 2026-11-12.