RuleSafe-VL - Decision State Prediction: leaderboard
Task 3 of RuleSafe-VL (Lu, Wang, Shang and Xu, Beijing University of Posts and Telecommunications, 2026), a benchmark of rule-conditioned content moderation built from platform policies on nudity, dangerous behavior and graphic injury: given a policy statement and an image-text case, the model says whether the visible evidence settles the outcome or the case is underdetermined because audience or purpose context is missing. Macro-F1 in percent over 2,166 expert-labelled cases, averaged over the three policy families; higher is better.
Metric: Macro-F1 (%; decidable versus underdetermined image-text cases, mean of three policy families). Source: arxiv.org. Saturation forecast: Around February 2028. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 58.6 | #100 |
| 2 | GPT-5.4 | 51.7 | #121 |
| 3 | Gemini 3.1 Pro (Preview) | 38.6 | #91 |
| 4 | Qwen 2.5 VL 7B Instruct | 34.6 | #1842 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=rulesafe-vl-decision-state-prediction · How It Works · Data refreshed daily, snapshot 2026-11-12.