WildFireVQA (RGB + Radiometric Statistics): leaderboard
WildFireVQA (2026) is an aerial wildfire-monitoring visual question answering benchmark built from paired RGB and radiometric thermal drone imagery, with multiple-choice answers generated by Gemini 2.5 Pro or, for some questions, computed from sensor data, then audited by manual verification and consistency checks. This board is accuracy in percent over all questions with RGB images plus a retrieved temperature summary of the paired thermal capture; higher is better.
Metric: Accuracy (%) over all WildFireVQA wildfire-monitoring multiple-choice questions (34 question types, 6,097 aerial frames each, from FLAME drone imagery with model-generated or sensor-derived labels, manually verified), RGB image input with a retrieved radiometric summary of the paired thermal TIFF (temperature statistics and pixel shares above 200 and 400 C) appended to the prompt; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 4 models tracked. Updated 15 Nov 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 8B Instruct | 54.76 | #1029 |
| 2 | llava-v1.6-mistral-7B | 51.88 | #3999 |
| 3 | InternVL2-8B | 47.66 | #3227 |
| 4 | MiniCPM-V-2 | 45.72 | #7564 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=wildfirevqa-rgb-plus-radiometric-statistics · How It Works · Data refreshed daily, snapshot 2026-11-21.