HoloCount - Visual-Prompt Region Grounding: 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 only inside a region marked on the image by a visual prompt such as a box (102 questions); higher is better). Source: arxiv.org. Saturation forecast: Around July 2028. 30 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 86.3 |
| 2 | Gemini 3 Flash (Preview) | 77.5 |
| 3 | Qwen 3.5 397B A17B | 76.5 |
| 4 | Qwen 3.5 122B A10B | 75.5 |
| 5 | Qwen 3.5 9B | 74.5 |
| 6 | Gemini 2.5 Pro | 71.6 |
| 7 | Qwen 3.5 35B A3B | 71.6 |
| 8 | Qwen 3.5 27B | 71.6 |
| 9 | Kimi K2.6 | 67.6 |
| 10 | Qwen 3.5 4B | 66.7 |
| 11 | Qwen 3 VL 32B Instruct | 65.7 |
| 12 | Qwen 2.5 VL 72B Instruct | 63.7 |
| 13 | GPT-4.1 | 61.8 |
| 14 | GPT-5.5 | 59.8 |
| 15 | Qwen 2.5 VL 32B Instruct | 58.8 |
Interactive version: theaggregate.ai/benchmark?slug=holocount-visual-prompt-region-grounding · How It Works · Data refreshed daily, snapshot 2026-09-29.