HoloCount - Action and State 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 in a stated action or state (102 questions); higher is better). Source: arxiv.org. Saturation forecast: Around December 2026. 30 models tracked.

Top models

#ModelScore
1Qwen 3.5 397B A17B89.2
2Qwen 3.5 122B A10B89.2
3Qwen 3.5 27B87.3
4Gemini 3.1 Pro (Preview)85.3
5Qwen 3.5 4B85.3
6GPT-5.584.3
7Qwen 3.5 35B A3B84.3
8Qwen 3.5 9B83.3
9Gemini 3 Flash (Preview)82.4
10Qwen 3 VL 32B Instruct82.4
11Gemini 2.5 Pro81.4
12Kimi K2.580.4
13Kimi K2.680.4
14Qwen 3 VL 8B Instruct78.4
15Claude Opus 4.777.5

Interactive version: theaggregate.ai/benchmark?slug=holocount-action-and-state-attribution · How It Works · Data refreshed daily, snapshot 2026-09-29.