K9-Bench - Cause-Effect Analysis: leaderboard

Metric: MCQ accuracy (%; the Cause-Effect Analysis category of K9-Bench, 4,744 five-option questions on 907 real-world dog videos; the free-form CoT answer counts as correct when its Qwen3-Embedding-8B similarity to the gold option exceeds 0.5 and no wrong option is closer; 32 uniformly sampled frames for GPT-4o and the open models, the full video at 1 FPS for Gemini 2.5 Pro and Qwen3-VL-235B-A22B). Source: arxiv.org. Saturation forecast: Around October 2027. 11 models tracked.

Top models

#ModelScore
1Gemini 2.5 Pro44.7
2Qwen 3 VL 235B A22B Instruct39.9
3InternVL3.5-8B36
4GPT-4o (2024-08-06)33.9
5Qwen 3 VL 32B (Thinking)32.8
6Qwen 3 VL 32B Instruct31.1
7Qwen 3 VL 4B Instruct28.2
8Qwen 3 VL 8B Instruct27.6
9Qwen 3 VL 8B (Thinking)24.9
10Qwen 3 VL 4B (Thinking)23.1

Interactive version: theaggregate.ai/benchmark?slug=k9-bench-cause-effect-analysis · How It Works · Data refreshed daily, snapshot 2026-09-29.