IMAVB: leaderboard
Omnimodal question answering over 500 long-form movie clips in which half of the questions carry a premise that contradicts the video or audio, testing whether a model flags the conflict instead of answering; IMAVB (Nguyen Quang et al., 2026). Balanced accuracy in percent (mean of standard and misleading vision and audio accuracy) on six-option questions with shuffled options, mean of three shuffles; higher is better.
Metric: Balanced accuracy (%; mean of standard vision, standard audio, misleading vision and misleading audio accuracy, 500 items each on 500 movie clips, six-option MCQ with options shuffled, mean of 3 shuffles). Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 61.5 | #91 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=imavb · How It Works · Data refreshed daily, snapshot 2026-11-12.