SpatialUAV - Region Recognition: leaderboard
Region recognition in real low-altitude UAV images from SpatialUAV (Zhang et al., 2026): the model names every labelled region that contains a queried object, with partial credit for a subset of the correct regions and zero when it names more regions than the answer holds. Mean per-question score in percent over 582 questions; higher is better.
Metric: Region Recognition Score (%; region-ID sets, partial credit, zero when more regions are named than exist). Source: arxiv.org. Saturation forecast: Around December 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 71.3 | #75 |
| 2 | Qwen 3.5 9B | 70 | #803 |
| 3 | Qwen 3.6 27B | 66.2 | #339 |
| 4 | GPT-5.4 | 56.7 | #121 |
| 5 | Qwen 3.6 35B A3B | 54.6 | #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-region-recognition · How It Works · Data refreshed daily, snapshot 2026-11-12.