SpatialUAV - Global Motion: leaderboard
UAV motion understanding from SpatialUAV (Zhang et al., 2026): from ordered video frames the model describes the camera and target motion in free text, and a GPT-5.4-mini judge scores its semantic similarity to the reference description from 0 to 1. Mean score in percent over 1,000 clips; higher is better.
Metric: Motion Description Similarity (%; GPT-5.4-mini judge, 0-1 per sample). Source: arxiv.org. Saturation forecast: Around June 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 | 66.6 | #121 |
| 2 | Claude Opus 4.7 | 51.9 | #75 |
| 3 | Qwen 3.6 27B | 50.5 | #339 |
| 4 | Qwen 3.5 9B | 48.4 | #803 |
| 5 | Qwen 3.6 35B A3B | 47.7 | #461 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=spatialuav-global-motion · How It Works · Data refreshed daily, snapshot 2026-11-12.