VSysBench - Aligned: leaderboard

Metric: Joint satisfaction rate (%; per-sample task score counted only when the judged constraint score reaches 0.8; aligned half, the user request compatible with the system constraint; 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

#ModelScore
1Claude Opus 4.742.9
2GPT-5.440.9
3Claude Sonnet 4.638
4GPT-5.4 Mini33.3
5Qwen 3 VL 32B29.7
6Claude Haiku 4.526.4
7GPT-4o24.2
8GPT-5.4 Nano22.2
9InternVL3.5-8B18.4
10Phi-4 Multimodal Instruct6.3

Interactive version: theaggregate.ai/benchmark?slug=vsysbench-aligned · How It Works · Data refreshed daily, snapshot 2026-09-29.