MMHBench - Third-Person: leaderboard
The third-person assessment questions of MMHBench (2026): 605 multiple-choice questions on long-form mental-health videos about observable behavior and multimodal evidence. Accuracy in percent; higher is better.
Metric: Accuracy (%; the 605 third-person assessment multiple-choice questions on 268 long-form mental-health-related videos, 64 uniformly sampled frames, zero-shot about observable behavior and multimodal evidence; higher is better). Source: arxiv.org. Saturation forecast: Around March 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 63.64 | #49 |
| 2 | Qwen 3.7 Plus | 58.35 | #207 |
| 3 | Qwen 3.6 Plus | 58.18 | #222 |
| 4 | Qwen 3.6 35B A3B | 57.85 | #462 |
| 5 | Qwen 3.5 35B A3B | 55.37 | #482 |
| 6 | Qwen 3.5 9B | 53.22 | #803 |
| 7 | Qwen 3.5 4B | 51.9 | #1175 |
| 8 | Gemma 4 26B A4B | 45.95 | #615 |
| 9 | MiniCPM-V-2.6 | 43.31 | #3063 |
| 10 | Ministral 3 14B | 30.08 | #1845 |
| 11 | Gemma 4 E4B | 28.6 | #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-third-person · How It Works · Data refreshed daily, snapshot 2026-11-13.