KnowHal - Knowledge (Negative): leaderboard
Metric: Accuracy (%; knowledge-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 2036. 14 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 2.5 VL 72B Instruct | 36.9 |
| 2 | Qwen 3 VL 32B Instruct | 35.6 |
| 3 | Gemini 3 Flash (Preview) | 27.6 |
| 4 | Kimi K2.5 | 25.5 |
| 5 | GPT-5.2 | 18.8 |
| 6 | Seed 2.0 Mini | 16.2 |
| 7 | InternVL3.5-8B | 13.1 |
| 8 | GPT-5.6 Sol | 11.9 |
| 9 | GLM-4.6V | 11.6 |
Interactive version: theaggregate.ai/benchmark?slug=knowhal-knowledge-negative · How It Works · Data refreshed daily, snapshot 2026-09-29.