HoloCount - Null-Target Prompting: leaderboard
Metric: Accuracy (%; exact-match accuracy of the single integer count a multimodal model returns zero-shot for an image and a counting question, with a strict integer-only system prompt; questions about objects absent from the image, where the correct count is zero (250 questions); higher is better). Source: arxiv.org. Saturation forecast: Estimated already saturated. 30 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3 VL 32B Instruct | 98 |
| 2 | Claude Sonnet 4.5 | 97.2 |
| 3 | Qwen 3 VL 8B Instruct | 96.4 |
| 4 | Claude Opus 4.7 | 95.6 |
| 5 | Kimi K2.6 | 95.6 |
| 6 | Qwen 2.5 VL 7B Instruct | 95.6 |
| 7 | Claude Sonnet 4.6 | 95.2 |
| 8 | Qwen 2.5 VL 72B Instruct | 95.2 |
| 9 | Claude Opus 4.6 | 94.4 |
| 10 | Kimi K2.5 | 94.4 |
| 11 | Qwen 2.5 VL 32B Instruct | 93.6 |
| 12 | Gemini 2.5 Pro | 93.4 |
| 13 | InternVL3.5-8B | 92.8 |
| 14 | Qwen 3.5 397B A17B | 90.8 |
| 15 | Qwen 3.5 122B A10B | 90.8 |
Interactive version: theaggregate.ai/benchmark?slug=holocount-null-target-prompting · How It Works · Data refreshed daily, snapshot 2026-09-29.