NormGym - Joint Success (Unseen Layouts): leaderboard
Interactive social-norm learning benchmark (arXiv:2611.08613): share of 150 unseen-layout transfer episodes across 50 hidden norms where a ReAct LLM/VLM agent completes both task goals with zero norm violations.
Metric: Joint Task & Norm-Compliance Success Rate (%), Unseen Layouts. Source: arxiv.org. Saturation forecast: Around February 2027. 7 models tracked. Updated 20 Nov 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 51.3 | #26 |
| 2 | GPT-5.6 Sol | 40.7 | #36 |
| 3 | Gemini 3.1 Pro (Preview) | 16.7 | #101 |
| 4 | Ministral-3-8B-Instruct | 4 | #1299 |
| 5 | Qwen 3 VL 8B Instruct | 0.6 | #1035 |
| 6 | Llama 3.2 11B Instruct | 0 | #5189 |
| 7 | GLM-4.6V-Flash | 0 | #1047 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=normgym-joint-success-unseen-layouts · How It Works · Data refreshed daily, snapshot 2026-11-20.