HoloCount - Material Attribution: 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 of a stated material (101 questions); higher is better). Source: arxiv.org. Saturation forecast: Around December 2026. 30 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.5 27B | 85.1 |
| 2 | Gemini 3.1 Pro (Preview) | 84.2 |
| 3 | Kimi K2.6 | 83.2 |
| 4 | Qwen 3.5 397B A17B | 83.2 |
| 5 | Qwen 3.5 35B A3B | 83.2 |
| 6 | Qwen 3.5 122B A10B | 82.2 |
| 7 | Qwen 3.5 9B | 80.2 |
| 8 | Gemini 3 Flash (Preview) | 77.2 |
| 9 | Qwen 3.5 4B | 77.2 |
| 10 | Gemini 2.5 Pro | 76.2 |
| 11 | GPT-5.5 | 76.2 |
| 12 | Kimi K2.5 | 75.2 |
| 13 | Claude Opus 4.8 | 75.2 |
| 14 | Qwen 2.5 VL 72B Instruct | 74.3 |
| 15 | Claude Opus 4.7 | 72.3 |
Interactive version: theaggregate.ai/benchmark?slug=holocount-material-attribution · How It Works · Data refreshed daily, snapshot 2026-09-29.