Obshazard-bench - Active Evolution Reasoning: leaderboard
The in-event stage of Obshazard-bench (National University of Defense Technology, Shanghai AI Laboratory and Shanghai Jiao Tong University, 2026): tracking an active disaster and predicting when it will end from raw Earth observation streams (213 samples). Task-aware soft score in percent, graded by distance to the true termination; higher is better.
Metric: Normalized task score (%; x100 of the 0-1 task-aware soft score following the Earth AI protocol: categorical answers by semantic or lexical consistency, numeric and interval answers by distance to the ground truth; equal-weight aggregation across timesteps, subtasks, lifecycle stages and disaster categories; zero-shot, uniform prompts; raw satellite sounding streams with ground-station, disaster-record and socio-economic inputs; in-situ disaster tracking and active termination prediction, 213 samples). Source: arxiv.org. Saturation forecast: Around 2029. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 20.78 | #48 |
| 2 | Kimi K2.6 | 14.88 | #173 |
| 3 | Qwen 3.5 397B A17B | 10.92 | #242 |
| 4 | Claude Opus 4.8 | 2.78 | #56 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=obshazard-bench-active-evolution-reasoning · How It Works · Data refreshed daily, snapshot 2026-11-12.