UAVQA-Bench - Existence Detection: leaderboard

Metric: Accuracy (%; scene, conditional and multi-target presence questions; 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 December 2026. 14 models tracked.

Top models

#ModelScore
1Gemini 3 Flash82.33
2Qwen 3 VL 32B Instruct80
3Gemini 3 Pro79.67
4Qwen 3 VL 8B (Thinking)79
5Qwen 3 VL 30B A3B (Thinking)78.33
6Qwen 3 VL 32B (Thinking)77.33
7Qwen 3 VL 30B A3B Instruct75
8Qwen 3 VL 8B Instruct72.33
9Qwen 3.5 9B65.67
10InternVL3.5-8B58

Interactive version: theaggregate.ai/benchmark?slug=uavqa-bench-existence-detection · How It Works · Data refreshed daily, snapshot 2026-09-29.