VERHallu - Counterfactual QA: leaderboard

Metric: Accuracy (%; 967 seven-option questions about an event absent from the clip, correct only when the model answers that the video information is incomplete; random 14.3; each model's own frame sampling and default settings, answer given as the option number). Source: arxiv.org. Saturation forecast: Around 2029. 18 models tracked.

Top models

#ModelScore
1Qwen 2.5 VL 72B Instruct67.6
2Qwen 2.5 VL 7B Instruct50.6
3Gemini 3 Pro45
4Qwen 2.5 VL 32B Instruct44.9
5MiniCPM-V-2.636.4
6InternVL3-8B15.6

Interactive version: theaggregate.ai/benchmark?slug=verhallu-counterfactual-qa · How It Works · Data refreshed daily, snapshot 2026-09-26.