PluRule: leaderboard
Metric: Accuracy (%; full context, 5,668 comments: each test instance pairs a rule-violating and a compliant comment; always answering 'no rules broken' scores 50). Source: arxiv.org. Saturation forecast: Around 2029. 8 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.2 (High) | 57.6 |
| 2 | GPT-5.2 (Low) | 57.4 |
| 3 | Qwen 3 VL 30B A3B Instruct | 52.3 |
| 4 | Qwen 3 VL 8B Instruct | 49.8 |
| 5 | Qwen 3 VL 30B A3B (Thinking) | 49.5 |
| 6 | Qwen 3 VL 4B Instruct | 48.4 |
| 7 | Qwen 3 VL 4B (Thinking) | 45 |
| 8 | Qwen 3 VL 8B (Thinking) | 44.9 |
Interactive version: theaggregate.ai/benchmark?slug=plurule · How It Works · Data refreshed daily, snapshot 2026-09-25.