GMP - Co-occurring Violations: leaderboard
Metric: Macro-F1 (x100) over the 12 violation categories of GMP Task A (identify every co-occurring violation in a post; 1,400 posts, 980 unsafe, 81% of them multi-label), zero-shot prompt with JSON output; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-4.5 (Non-reasoning) | 58 | #265 (GLM-4.5) |
| 2 | Claude Sonnet 4 | 57.58 | #194 |
| 3 | DeepSeek V3.1 (Non-reasoning) | 57.37 | #260 (DeepSeek V3.1) |
| 4 | Qwen 2.5 VL 72B Instruct | 54.94 | #364 |
| 5 | GPT-4o Mini | 53.72 | #588 |
| 6 | GLM-4.5 (Thinking) | 52 | #265 (GLM-4.5) |
| 7 | Gemini 2.5 Flash | 51.87 | #237 |
| 8 | Qwen 3 30B A3B 2507 Instruct | 50.71 | #464 |
| 9 | Kimi K2 | 49 | #236 |
| 10 | DeepSeek V3.1 (Thinking) | 47 | #260 (DeepSeek V3.1) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=gmp-co-occurring-violations · How It Works · Data refreshed daily, snapshot 2026-10-11.