MoHallBench - Sequential Inference Hallucination: leaderboard
Metric: Q-PairAcc (%; share of adversarial pairs in which the model answers both the positive and the negative (reversed) yes/no query correctly; 32 uniformly sampled frames, temperature 0; sequential inference hallucination on clips truncated before the terminal sub-action). Source: arxiv.org. Saturation forecast: Around September 2028. 10 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Qwen 3 VL 8B Instruct | 11.5 |
| 2 | Molmo2-8B | 11.1 |
| 3 | Qwen 2.5 VL 7B Instruct | 11 |
| 4 | Qwen 2 VL 7B Instruct | 7.1 |
| 5 | InternVL3-8B | 5.5 |
Interactive version: theaggregate.ai/benchmark?slug=mohallbench-sequential-inference-hallucination · How It Works · Data refreshed daily, snapshot 2026-09-29.