MMJailBench - Economic Harm: leaderboard
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 |
|---|---|---|
| 1 | GPT-5 | 4.11 |
| 2 | Claude Sonnet 4.5 | 15.11 |
| 3 | Qwen 3 VL 8B | 30.28 |
| 4 | Gemini 3 Flash | 38.94 |
| 5 | Kimi K2.5 | 45.11 |
| 6 | Grok 4 Fast | 52.5 |
| 7 | InternVL3-8B | 61.72 |
| 8 | Seed 2.0 Lite | 61.94 |
| 9 | Ministral 3 8B | 63.78 |
| 10 | Step3 VL 10B | 68.22 |
| 11 | Qwen 2.5 VL 7B Instruct | 70.83 |
| 12 | Gemma 3 12B | 80.44 |
| 13 | GLM-4.6V | 88.11 |
Interactive version: theaggregate.ai/benchmark?slug=mmjailbench-economic-harm · How It Works · Data refreshed daily, snapshot 2026-09-26.