VSysBench - Constraint Satisfaction: leaderboard

Metric: Constraint satisfaction rate (%; share of responses whose judged constraint score reaches 0.8, both halves; 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 December 2026. 16 models tracked.

Top models

#ModelScore
1GPT-5.483.4
2Claude Opus 4.777.2
3GPT-5.4 Mini71.9
4Claude Sonnet 4.669.7
5GPT-5.4 Nano63.2
6Claude Haiku 4.559.1
7GPT-4o41.4
8Qwen 3 VL 32B34.6
9InternVL3.5-8B33.8
10Phi-4 Multimodal Instruct19

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