HoloCount - Relative Canonical Orientation: 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 inside a region named by a canonical direction such as the bottom-left quadrant (84 questions); higher is better). Source: arxiv.org. Saturation forecast: Around August 2027. 30 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3.5 397B A17B | 82.1 |
| 2 | Qwen 3.5 27B | 82.1 |
| 3 | Gemini 3.1 Pro (Preview) | 81 |
| 4 | Qwen 3.5 35B A3B | 78.6 |
| 5 | Gemini 3 Flash (Preview) | 78.6 |
| 6 | Qwen 3.5 122B A10B | 77.4 |
| 7 | Kimi K2.5 | 76.2 |
| 8 | Qwen 3.5 9B | 76.2 |
| 9 | Kimi K2.6 | 76.2 |
| 10 | GPT-5.5 | 73.8 |
| 11 | Qwen 3 VL 32B Instruct | 72.6 |
| 12 | Qwen 3.5 4B | 71.4 |
| 13 | Qwen 2.5 VL 32B Instruct | 71.4 |
| 14 | Claude Opus 4.7 | 69 |
| 15 | Claude Sonnet 4.6 | 67.9 |
Interactive version: theaggregate.ai/benchmark?slug=holocount-relative-canonical-orientation · How It Works · Data refreshed daily, snapshot 2026-09-29.