MHDash: leaderboard
Metric: Average macro-F1 (%, the mean of the concern-type (7 classes) and risk-level (6 classes) macro-F1, printed as fractions and shown times 100) on the MHDash test split (150 of 1,000 GPT-4o-generated ten-round support dialogues, each conditioned on an expert-annotated social-media post whose concern-type and risk-level labels it inherits; 56.9% of dialogues are 'not related' on both dimensions); five-shot prompting with the same prompt for every model, no fine-tuning; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o Mini | 23.23 | #588 |
| 2 | GPT-4o | 22.65 | #333 |
| 3 | GPT-3.5 | 21.58 | #809 |
| 4 | DeepSeek V3 | 21.34 | #312 |
| 5 | Llama 3.1 70B Instruct | 20.67 | #548 |
| 6 | Llama 3.3 70B Instruct | 19.02 | #520 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=mhdash · How It Works · Data refreshed daily, snapshot 2026-10-11.