SafeSceneReason - Spatial and Structural Reasoning: leaderboard

Metric: Accuracy (%; the 164 spatial/structural reasoning 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 Sol86.59
2GPT-5.584.15
3Gemini 3.5 Flash76.22
4Gemini 3.1 Pro (Preview)75
5Claude Fable 573.17
6Kimi K2.665.24
7Qwen 3.6 27B51.22
8Qwen 3.5 9B45.73

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