MotionHalluc: leaderboard
Closed-form questions on directional, attributional and temporal motion differences between two videos (1,540 questions). MotionHalluc (Guo et al., 2026) tests whether multimodal models compare a query and a reference fitness-exercise video (553 pairs from Fit3D) without hallucinating motion differences; open models decode greedily and API models are averaged over three runs. The score is the unweighted mean of the three task accuracies in percent; higher is better.
Metric: Accuracy (%; unweighted mean of the directional, attributional and temporal task accuracies; 1,540 questions over 553 paired fitness videos). Source: arxiv.org. Saturation forecast: Around February 2027. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 Plus | 71.28 | #227 |
| 2 | Gemini 3 Flash | 67.21 | #159 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=motionhalluc · How It Works · Data refreshed daily, snapshot 2026-11-12.