DermoBench - Hierarchical Diagnosis: leaderboard
Step-wise dermatology diagnosis from DermoBench (DermoGPT paper, Ru et al., 2026): for 2,000 held-out images the model descends a 325-class diagnostic ontology from superclass to leaf through a sequence of multiple-choice decisions. Accuracy in percent; higher is better.
Metric: Accuracy (%; step-wise root-to-leaf diagnosis along the ontology, in-distribution). Source: arxiv.org. Saturation forecast: Around February 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 70.31 | #452 |
| 2 | GPT-4o Mini | 65.9 | #1627 |
| 3 | Claude Sonnet 4.5 (Thinking) | 63.4 | #244 (Claude Sonnet 4.5) |
| 4 | QVQ-72B-Preview | 57.25 | #1173 |
| 5 | Qwen 2.5 VL 72B Instruct | 53.93 | #845 |
| 6 | GLM-4.5V | 52.39 | #739 |
| 7 | Llama 3.2 90B | 51.2 | #1740 |
| 8 | Qwen 3 VL 8B Instruct | 44.77 | #982 |
| 9 | Nemotron Nano 12B v2 VL | 40.17 | #2248 |
| 10 | Llama 3.2 11B | 35.98 | #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-hierarchical-diagnosis · How It Works · Data refreshed daily, snapshot 2026-11-12.