KnowHal - Attribute (Negative): leaderboard

Metric: Accuracy (%; attribute-dimension negative (false-premise) 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 2030. 14 models tracked.

Top models

#ModelScore
1Qwen 2.5 VL 72B Instruct72.3
2Qwen 3 VL 32B Instruct70.9
3Gemini 3 Flash (Preview)57.6
4Kimi K2.552.8
5GPT-5.250.6
6InternVL3.5-8B50.1
7Seed 2.0 Mini49.7
8GLM-4.6V45.2
9GPT-5.6 Sol35.9

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