KnowHal - Relation (Positive): leaderboard

Metric: Accuracy (%; relation-dimension positive questions; 1,800 entity-image samples in 10 domains with paired positive and false-premise negative questions; positive answers scored by standard VQA answer normalization and soft matching, negative answers by a GPT-4o-mini judge (option matching for relation questions); zero-shot). Source: arxiv.org. Saturation forecast: Around December 2026. 14 models tracked.

Top models

#ModelScore
1Gemini 3 Flash (Preview)89.6
2GLM-4.6V86.8
3Seed 2.0 Mini86.2
4GPT-5.286.1
5GPT-5.6 Sol84.7
6Qwen 2.5 VL 72B Instruct82.9
7Qwen 3 VL 32B Instruct82.3
8Kimi K2.581.9
9InternVL3.5-8B80.2

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