SafeSceneReason - Hazard Identification: leaderboard

Metric: Accuracy (%; the 84 hazard identification questions; single-answer multiple choice on the 1,000-question test set balanced across answer positions, built from workplace scene graphs and NIOSH FACE and CSB accident-report figures; temperature 0, unparseable answers counted wrong). Source: arxiv.org. Saturation forecast: Around December 2026. 8 models tracked.

Top models

#ModelScore
1GPT-5.6 Sol89.29
2GPT-5.585.71
3Claude Fable 577.38
4Gemini 3.1 Pro (Preview)73.81
5Gemini 3.5 Flash73.81
6Kimi K2.653.57
7Qwen 3.5 9B48.81
8Qwen 3.6 27B42.86

Interactive version: theaggregate.ai/benchmark?slug=safescenereason-hazard-identification · How It Works · Data refreshed daily, snapshot 2026-09-26.