HoloCount - Small-Scale Enumeration: 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; counting objects smaller than 32 by 32 pixels (101 questions); higher is better). Source: arxiv.org. Saturation forecast: Around 2029. 30 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 74.3 |
| 2 | Gemini 3.1 Pro (Preview) | 73.3 |
| 3 | Qwen 3.5 397B A17B | 60.4 |
| 4 | Qwen 3.5 35B A3B | 60.4 |
| 5 | Qwen 3.5 122B A10B | 58.4 |
| 6 | Qwen 3.5 27B | 52.5 |
| 7 | Qwen 3.5 9B | 50.5 |
| 8 | Qwen 2.5 VL 72B Instruct | 43.6 |
| 9 | GPT-5.5 | 42.6 |
| 10 | Qwen 3.5 4B | 42.6 |
| 11 | Claude Opus 4.7 | 40.6 |
| 12 | Kimi K2.6 | 40.6 |
| 13 | Gemini 2.5 Pro | 39.6 |
| 14 | Claude Opus 4.8 | 38.6 |
| 15 | Kimi K2.5 | 37.6 |
Interactive version: theaggregate.ai/benchmark?slug=holocount-small-scale-enumeration · How It Works · Data refreshed daily, snapshot 2026-09-29.