MMHBench - First-Person: leaderboard
The first-person perspective-taking questions of MMHBench (2026): 1,579 multiple-choice questions on long-form mental-health videos, asked in the voice of a social role such as the person with lived experience, a clinician or a family member. Accuracy in percent; higher is better.
Metric: Accuracy (%; the 1,579 first-person perspective-taking multiple-choice questions on 268 long-form mental-health-related videos, 64 uniformly sampled frames, zero-shot, asked in the voice of a social role such as the person with lived experience, a mental health professional or a family member; higher is better). Source: arxiv.org. Saturation forecast: Around September 2028. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 35B A3B | 53.7 | #462 |
| 2 | GPT-5.5 | 52.12 | #49 |
| 3 | Qwen 3.5 9B | 50.73 | #803 |
| 4 | Qwen 3.5 35B A3B | 50.35 | #482 |
| 5 | Qwen 3.7 Plus | 48.89 | #207 |
| 6 | Qwen 3.5 4B | 48.07 | #1175 |
| 7 | Qwen 3.6 Plus | 45.79 | #222 |
| 8 | Gemma 4 26B A4B | 44.65 | #615 |
| 9 | MiniCPM-V-2.6 | 35.53 | #3063 |
| 10 | Ministral 3 14B | 30.4 | #1845 |
| 11 | Gemma 4 E4B | 28.82 | #2214 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=mmhbench-first-person · How It Works · Data refreshed daily, snapshot 2026-11-13.