RuleSafe-VL - Atomic Rule Identification: leaderboard
Task 1 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 an image-text case and the candidate atomic rules of its policy, the vision-language model says which rules the case satisfies. Macro-F1 in percent, averaged over the three policy families; higher is better.
Metric: Macro-F1 (%; whether each candidate atomic policy rule is satisfied by an image-text case, mean of three policy families). Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.6 | 69.3 | #100 |
| 2 | Gemini 3.1 Pro (Preview) | 68.8 | #91 |
| 3 | GPT-5.4 | 56.8 | #121 |
| 4 | Qwen 2.5 VL 7B Instruct | 47 | #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-atomic-rule-identification · How It Works · Data refreshed daily, snapshot 2026-11-12.