MMJailBench - System-Role Framing: leaderboard
Metric: Attack success rate (%; share of responses a GPT-5 judge scores 4 or 5 on a 1-5 harmfulness rubric; system-role simulation 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 | 0 |
| 2 | Claude Sonnet 4.5 | 0 |
| 3 | InternVL3-8B | 0.07 |
| 4 | Qwen 2.5 VL 7B Instruct | 0.66 |
| 5 | Gemini 3 Flash | 3.2 |
| 6 | Grok 4 Fast | 3.53 |
| 7 | Step3 VL 10B | 15.51 |
| 8 | Seed 2.0 Lite | 16.1 |
| 9 | Gemma 3 12B | 18.9 |
| 10 | Qwen 3 VL 8B | 22.32 |
| 11 | Ministral 3 8B | 30.7 |
| 12 | Kimi K2.5 | 40.33 |
| 13 | GLM-4.6V | 79.56 |
Interactive version: theaggregate.ai/benchmark?slug=mmjailbench-system-role-framing · How It Works · Data refreshed daily, snapshot 2026-09-26.