FinSafetyBench (ReNeLLM) - Ethical Violations (Chinese): leaderboard
Metric: Attack success rate (%; ReNeLLM jailbreak; share of the benchmark's ethical-violation instructions in Chinese whose jailbroken response a Qwen3-32B judge rates a complete violation (10 of 10)). Source: arxiv.org. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Llama 3 8B Instruct | 33.97 |
| 2 | InternLM3-8B-Instruct | 44.23 |
| 3 | Qwen 2.5 32B Instruct | 60.16 |
| 4 | Mistral Small 3 | 70.64 |
| 5 | GLM-4 9B (0414) | 73.21 |
Interactive version: theaggregate.ai/benchmark?slug=finsafetybench-renellm-ethical-violations-chinese · How It Works · Data refreshed daily, snapshot 2026-09-26.