UAVQA-Bench - Spatial Relationship Understanding: leaderboard
UAVQA-Bench (Fudan University, 2026) asks multimodal models 1,500 human-annotated questions about drone images drawn from 13 public UAV datasets, mostly multiple choice with some bounding-box grounding. This dimension covers spatial relations, height comparison and distance comparison from the aerial view. Score is accuracy in percent; higher is better.
Metric: Accuracy (%; spatial relations, height comparison and distance comparison from the aerial view; direct answering over UAVQA-Bench: 1,500 human-annotated questions on images from 13 public UAV datasets, 16 tasks in six capability dimensions, multiple choice plus box grounding counted correct at IoU 0.5). Source: arxiv.org. Saturation forecast: Around February 2027. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 60.8 | #137 |
| 2 | Qwen 3 VL 32B Instruct | 59.6 | #585 |
| 3 | Qwen 3 VL 32B (Thinking) | 58.4 | #559 |
| 4 | Gemini 3 Flash | 56.4 | #164 |
| 5 | Qwen 3 VL 30B A3B (Thinking) | 56 | #712 |
| 6 | Qwen 3.5 9B | 50.8 | #803 |
| 7 | Qwen 3 VL 30B A3B Instruct | 43.2 | #871 |
| 8 | Qwen 3 VL 8B Instruct | 41.6 | #981 |
| 9 | Qwen 3 VL 8B (Thinking) | 39.6 | #887 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=uavqa-bench-spatial-relationship-understanding · How It Works · Data refreshed daily, snapshot 2026-11-13.