DisasterBench (UAV): leaderboard
DisasterBench (2026, UAV disaster response) is a multiple-choice reasoning benchmark over low-altitude drone images of 14 disaster-related scene types, covering tasks before, during and after a disaster. The headline is exact-match accuracy (%) over the nine tasks of the 3,000-question test split; higher is better.
Metric: Overall accuracy (%) over the nine response tasks on the 3,000-question test split of low-altitude UAV disaster images (14 disaster scene types), multiple choice scored by strict exact match on the final option letter; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 84.83 | #167 |
| 2 | Gemini 2.5 Pro | 84.17 | #266 |
| 3 | Claude Sonnet 4.5 | 79.93 | #250 |
| 4 | Gemini 2.0 Flash | 78.8 | #704 |
| 5 | Gemini 2.5 Flash | 78.03 | #453 |
| 6 | Grok 4 | 77.8 | #307 |
| 7 | GPT-4o | 73 | #741 |
| 8 | Qwen 2.5 VL 7B Instruct | 71.6 | #1833 |
| 9 | InternVL2-8B | 68.23 | #2988 |
| 10 | GPT-4o Mini | 64 | #1623 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=disasterbench-uav · How It Works · Data refreshed daily, snapshot 2026-11-13.