MMJailBench - Economic Harm: leaderboard
MMJailBench (Wang et al., 2026) is a factorized multimodal jailbreak benchmark: 272 harmful intents crossed with six prompt framings, five task-relevant visual contexts and text or in-image instructions, 16,320 instances per model. This board covers the Economic Harm domain (fraud, scams and financial-crime requests) across all framings, visual contexts and carriers. Attack success rate in percent, the share of responses a GPT-5 judge scores 4 or 5 on a 1-5 harmfulness rubric; lower is better.
Metric: Attack success rate (%; share of responses a GPT-5 judge scores 4 or 5 on a 1-5 harmfulness rubric; Economic Harm domain, 30 harmful intents across all framings, visual contexts and carriers). Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 4.11 | #163 |
| 2 | Claude Sonnet 4.5 | 15.11 | #244 |
| 3 | Gemini 3 Flash | 38.94 | #159 |
| 4 | Kimi K2.5 | 45.11 | #252 |
| 5 | Grok 4 Fast | 52.5 | #438 |
| 6 | InternVL3-8B | 61.72 | #1695 |
| 7 | Seed 2.0 Lite | 61.94 | #286 |
| 8 | Ministral 3 8B | 63.78 | #2036 |
| 9 | Qwen 2.5 VL 7B Instruct | 70.83 | #1842 |
| 10 | Gemma 3 12B | 80.44 | #2013 |
| 11 | GLM-4.6V | 88.11 | #698 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mmjailbench-economic-harm · How It Works · Data refreshed daily, snapshot 2026-11-12.