CIBuzzBench - Harmfulness Detection (Chinese Prompt): leaderboard
CIBuzzBench (Wang et al., 2026) tests cross-lingual understanding of Chinese internet buzzwords. In Harmfulness Detection the model classifies the documented online sense of a buzzword as harmful or non-harmful from the buzzword alone under Chinese-language instructions, with the Chinese buzzword as input; macro F1 in percent, higher is better.
Metric: Macro F1 (%; binary harmful or non-harmful classification of the documented online sense from the buzzword alone, 14.2% of entries harmful; Chinese instructions; 3,001 Chinese internet buzzwords; zero-shot, temperature 0; the Chinese buzzword is the input under both prompt languages). Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 79.86 | #91 |
| 2 | GLM-5.2 | 78.39 | #137 |
| 3 | GPT-5.5 | 77.41 | #48 |
| 4 | Qwen 3.7 Max | 76.1 | #142 |
| 5 | DeepSeek V4 Pro | 74.68 | #181 |
| 6 | Claude Opus 4.8 | 70.41 | #56 |
| 7 | Qwen 3 14B | 63.31 | #1235 |
| 8 | Qwen 3 8B | 62.55 | #1838 |
| 9 | Qwen 3 4B | 22.83 | #2297 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=cibuzzbench-harmfulness-detection-chinese-prompt · How It Works · Data refreshed daily, snapshot 2026-11-12.