DermoBench - Diagnosis (25-Way Triage): leaderboard
Triage-style dermatology diagnosis from DermoBench (DermoGPT paper, Ru et al., 2026): each of 2,000 held-out images must be placed in one of 25 coarse diagnostic categories collapsed from a 325-class ontology, as a 25-option multiple-choice question. Accuracy in percent; higher is better.
Metric: Accuracy (%; 25-way coarse-category MCQA, in-distribution). Source: arxiv.org. Saturation forecast: Around October 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.2 90B | 51.65 | #1740 |
| 2 | QVQ-72B-Preview | 47.3 | #1173 |
| 3 | Gemini 2.5 Flash | 47.2 | #452 |
| 4 | Qwen 3 VL 8B Instruct | 45.35 | #982 |
| 5 | Qwen 2.5 VL 72B Instruct | 35.95 | #845 |
| 6 | GPT-4o Mini | 34.75 | #1627 |
| 7 | Claude Sonnet 4.5 (Thinking) | 34.15 | #244 (Claude Sonnet 4.5) |
| 8 | GLM-4.5V | 28.85 | #739 |
| 9 | Nemotron Nano 12B v2 VL | 25.6 | #2248 |
| 10 | Llama 3.2 11B | 16.5 | #6434 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=dermobench-diagnosis-25-way-triage · How It Works · Data refreshed daily, snapshot 2026-11-12.