TF-RefusalBench - Summarization Refusal: leaderboard

Metric: Refusal rate (%; share of 2,800 summarization prompts over Swiss Federal Supreme Court criminal-law passages in which the model declines the task for content or ethical reasons; 3 samples per prompt at temperature 0.7, 2-of-3 majority of three LLM judges). Source: arxiv.org. Saturation forecast: Estimated already saturated. 5 models tracked.

Top models

#ModelScore
1Gemma 4 31B (IT)0
2Qwen 3 32B0
3Llama 3.3 70B Instruct7.2
4GPT-OSS-20B14.7

Interactive version: theaggregate.ai/benchmark?slug=tf-refusalbench-summarization-refusal · How It Works · Data refreshed daily, snapshot 2026-09-26.