MMJailBench - Safety-Paradox Framing: leaderboard
Metric: Attack success rate (%; share of responses a GPT-5 judge scores 4 or 5 on a 1-5 harmfulness rubric; safety-paradox reasoning framing, all 272 intents, visual contexts and carriers). Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5 | 1.88 |
| 2 | Claude Sonnet 4.5 | 3.27 |
| 3 | Qwen 3 VL 8B | 13.49 |
| 4 | Kimi K2.5 | 15.15 |
| 5 | Gemini 3 Flash | 18.97 |
| 6 | Ministral 3 8B | 28.24 |
| 7 | GLM-4.6V | 33.71 |
| 8 | Qwen 2.5 VL 7B Instruct | 35.62 |
| 9 | Gemma 3 12B | 37.06 |
| 10 | InternVL3-8B | 43.09 |
| 11 | Grok 4 Fast | 43.12 |
| 12 | Step3 VL 10B | 48.42 |
| 13 | Seed 2.0 Lite | 51.95 |
Interactive version: theaggregate.ai/benchmark?slug=mmjailbench-safety-paradox-framing · How It Works · Data refreshed daily, snapshot 2026-09-26.