MoHallBench - Semantic Similarity Hallucination (Multiple Choice): leaderboard
Four-option questions contrasting the true action with a semantically related distractor, an unrelated action and a None option; from MoHallBench (2026). Accuracy in percent, higher is better.
Metric: Multiple-choice accuracy (%; four-option multiple choice (ground truth, hard hallucination distractor, random distractor, None of these); 32 uniformly sampled frames, temperature 0; semantic similarity hallucination). Source: arxiv.org. Saturation forecast: Estimated already saturated. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 8B Instruct | 81 | #981 |
| 2 | Qwen 2.5 VL 7B Instruct | 79.3 | #1833 |
| 3 | InternVL3-8B | 78.9 | #1688 |
| 4 | Qwen 2 VL 7B Instruct | 78.7 | #2553 |
| 5 | Molmo2-8B | 77.3 | #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-semantic-similarity-hallucination-multiple-choice · How It Works · Data refreshed daily, snapshot 2026-11-13.