MMHBench: leaderboard
MMHBench (Hefei University of Technology and others, 2026) evaluates multimodal LLMs on mental health understanding in 268 long-form videos with 2,184 multiple-choice questions, zero-shot with 64 sampled frames. Accuracy in percent over all questions; higher is better.
Metric: Overall accuracy (%; all 2,184 multiple-choice questions on 268 long-form mental-health-related videos, 64 uniformly sampled frames, zero-shot; higher is better). Source: arxiv.org. Saturation forecast: Around March 2028. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 55.31 | #49 |
| 2 | Qwen 3.6 35B A3B | 54.85 | #462 |
| 3 | Qwen 3.5 35B A3B | 51.74 | #482 |
| 4 | Qwen 3.7 Plus | 51.51 | #207 |
| 5 | Qwen 3.5 9B | 51.42 | #803 |
| 6 | Qwen 3.6 Plus | 49.22 | #222 |
| 7 | Qwen 3.5 4B | 49.13 | #1175 |
| 8 | Gemma 4 26B A4B | 45.01 | #615 |
| 9 | MiniCPM-V-2.6 | 37.68 | #3063 |
| 10 | Ministral 3 14B | 30.31 | #1845 |
| 11 | Gemma 4 E4B | 28.75 | #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 · How It Works · Data refreshed daily, snapshot 2026-11-13.