HoloCount - High-Density 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 densely packed synthetic symbols on a blank canvas (100 questions); higher is better). Source: arxiv.org. Saturation forecast: Around February 2028. 30 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 49 |
| 2 | GPT-5.5 | 20 |
| 3 | Qwen 3.5 27B | 20 |
| 4 | Qwen 3.5 397B A17B | 14 |
| 5 | Gemini 3 Flash (Preview) | 12 |
| 6 | Qwen 3.5 122B A10B | 9 |
| 7 | GPT-5.4 | 8 |
| 8 | Qwen 3.5 35B A3B | 5 |
| 9 | Claude Opus 4.7 | 4 |
| 10 | Kimi K2.6 | 4 |
| 11 | InternVL3.5-8B | 4 |
| 12 | Kimi K2.5 | 3 |
| 13 | Qwen 3.5 9B | 3 |
| 14 | Qwen 3 VL 32B Instruct | 3 |
| 15 | Claude Sonnet 4.5 | 2 |
Interactive version: theaggregate.ai/benchmark?slug=holocount-high-density-enumeration · How It Works · Data refreshed daily, snapshot 2026-09-29.