VSysBench - Misaligned: leaderboard
Metric: Joint satisfaction rate (%; per-sample task score counted only when the judged constraint score reaches 0.8; misaligned half, the user explicitly contradicts the system message with a direct jailbreak prefix; VSysBench: 2,258 verified MM-Vet v2 image questions, each under a system-message constraint from 22 sub-categories in 5 categories, once with an aligned and once with a conflicting user message (4,516 instances); default inference settings; GPT-5-mini judge at temperature 0). Source: arxiv.org. Saturation forecast: Around October 2027. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.4 | 31.5 |
| 2 | Claude Opus 4.7 | 29.5 |
| 3 | Claude Sonnet 4.6 | 25.8 |
| 4 | GPT-5.4 Mini | 23.2 |
| 5 | Claude Haiku 4.5 | 17.6 |
| 6 | GPT-5.4 Nano | 16.1 |
| 7 | GPT-4o | 11.1 |
| 8 | InternVL3.5-8B | 7 |
| 9 | Qwen 3 VL 32B | 3.6 |
| 10 | Phi-4 Multimodal Instruct | 2.6 |
Interactive version: theaggregate.ai/benchmark?slug=vsysbench-misaligned · How It Works · Data refreshed daily, snapshot 2026-09-29.