XWOD - Weather Classification: leaderboard

Metric: Top-1 accuracy (%, times 100) of XWOD-LLM-WC: a vision-language model names the dominant extreme weather of a single street-level image as one of seven labels (rain, snow, fog, haze/sand/dust, flooding, tornado, wildfire), on a class-balanced probe of 700 XWOD test images (100 per class, fixed seed, longest side at most 1024 px); one zero-shot single-turn prompt without chain of thought, the free-text reply canonicalized by exact match and a keyword fallback, unmatched replies counted wrong; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 4 models tracked.

Top models

Interactive version: theaggregate.ai/benchmark?slug=xwod-weather-classification · How It Works · Data refreshed daily, snapshot 2026-10-07.