MoHallBench - Visual Similarity Hallucination (Multiple Choice): leaderboard
Four-option questions on a positive clip or its visually near-identical negative clip, where the negative expects None of these; from MoHallBench (2026). Accuracy in percent, higher is better.
Metric: Multiple-choice accuracy (%; dual-video four-option choice where the negative clip expects None of these; 32 uniformly sampled frames, temperature 0; visual similarity hallucination). Source: arxiv.org. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 8B Instruct | 59.2 | #981 |
| 2 | InternVL3-8B | 58.3 | #1688 |
| 3 | Qwen 2.5 VL 7B Instruct | 56.6 | #1833 |
| 4 | Qwen 2 VL 7B Instruct | 52 | #2553 |
| 5 | Molmo2-8B | 51.2 | #1483 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mohallbench-visual-similarity-hallucination-multiple-choice · How It Works · Data refreshed daily, snapshot 2026-11-13.