CringeBench: leaderboard

Metric: Cringe Score (0-10, lower is better). Source: huggingface.co. 34 models tracked.

Top models

#ModelScore
1GPT-5.20.29
2Claude Sonnet 4.50.54
3MiniMax-M2.50.54
4Molmo2-8B0.63
5OLMo 3 7B Instruct0.69
6Devstral Small 20.74
7trinity-large-preview0.76
8Qwen 3 235B A22B 25070.76
9Step 3.5 Flash0.83
10DeepSeek V3.20.84
11GLM-4.50.85
12Llama 3.3 70B Instruct0.92
13Llama 4 Maverick0.92
14Claude Opus 4.60.92
15GLM-4.60.95

Interactive version: theaggregate.ai/benchmark?slug=cringebench · How It Works · Data refreshed daily, snapshot 2026-09-05.