CIBuzzBench - Harmfulness Detection (English Prompt): leaderboard

Metric: Macro F1 (%; binary harmful or non-harmful classification of the documented online sense from the buzzword alone, 14.2% of entries harmful; English 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 December 2026. 9 models tracked.

Top models

#ModelScore
1Gemini 3.1 Pro (Preview)80.45
2GPT-5.576.23
3DeepSeek V4 Pro75.36
4Qwen 3.7 Max74
5GLM-5.272.94
6Claude Opus 4.871.09
7Qwen 3 14B59.09
8Qwen 3 8B58.62
9Qwen 3 4B28.77

Interactive version: theaggregate.ai/benchmark?slug=cibuzzbench-harmfulness-detection-english-prompt · How It Works · Data refreshed daily, snapshot 2026-09-26.