MMOOC - Visual False Premises (Yes/No): leaderboard

Metric: Out-of-context score (%; mean of the refusal rate, the share of the Visual False Premises Yes/No questions, which the image and context cannot answer, that the model correctly declines, and refusal rationality, the mean 0-1 judge score of its abstention reasoning in steps of 0.25; judge scores averaged over GPT-5.6, Claude Opus 5 and DeepSeek-V4-Pro). Source: arxiv.org. Saturation forecast: Around 2029. 18 models tracked.

Top models

#ModelScore
1O171.75
2Claude Opus 4.667.75
3Gemma 4 26B66.25
4Qwen 3.5 122B A10B65.25
5Qwen 3.5 27B63.25
6Ministral 3 8B60.5
7GPT-4o59.25
8InternVL3-8B55.5
9Gemma 4 31B54.25
10Ministral 3 14B52.75
11Llama 4 Maverick51.25
12O335.5
13Gemini 3.1 Pro (Preview)9

Interactive version: theaggregate.ai/benchmark?slug=mmooc-visual-false-premises-yes-no · How It Works · Data refreshed daily, snapshot 2026-09-29.