HoloCount - Differential Comparison: 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; the difference between the counts of two object sets (100 questions); higher is better). Source: arxiv.org. Saturation forecast: Around December 2026. 30 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 79 |
| 2 | Gemini 3 Flash (Preview) | 78 |
| 3 | Qwen 3.5 27B | 76 |
| 4 | Qwen 3.5 397B A17B | 74 |
| 5 | Kimi K2.6 | 70 |
| 6 | Qwen 3.5 122B A10B | 70 |
| 7 | Qwen 3.5 35B A3B | 69 |
| 8 | Qwen 3.5 9B | 67 |
| 9 | Qwen 3.5 4B | 67 |
| 10 | Kimi K2.5 | 66 |
| 11 | Gemini 2.5 Pro | 65 |
| 12 | Claude Opus 4.8 | 59 |
| 13 | GPT-5.5 | 58 |
| 14 | Claude Opus 4.7 | 58 |
| 15 | MiniMax-M3 | 56 |
Interactive version: theaggregate.ai/benchmark?slug=holocount-differential-comparison · How It Works · Data refreshed daily, snapshot 2026-09-29.