DermoBench - OOD Diagnosis - Derm1M: leaderboard
Out-of-distribution skin-lesion diagnosis on the Derm1M educational split, part of DermoBench (DermoGPT paper, Ru et al., 2026): four-option multiple-choice questions written in the dataset's own label space rather than the benchmark's unified ontology. Accuracy in percent; higher is better.
Metric: Accuracy (%; 4-way MCQA in the source dataset's own label space, out-of-distribution). Source: arxiv.org. Saturation forecast: Around February 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Flash | 66.33 | #452 |
| 2 | QVQ-72B-Preview | 60.53 | #1173 |
| 3 | Qwen 2.5 VL 72B Instruct | 54.63 | #845 |
| 4 | Claude Sonnet 4.5 (Thinking) | 53.64 | #244 (Claude Sonnet 4.5) |
| 5 | Qwen 3 VL 8B Instruct | 52.67 | #982 |
| 6 | GPT-4o Mini | 52.12 | #1627 |
| 7 | GLM-4.5V | 45.51 | #739 |
| 8 | Llama 3.2 90B | 44.76 | #1740 |
| 9 | Nemotron Nano 12B v2 VL | 44.12 | #2248 |
| 10 | Llama 3.2 11B | 25.5 | #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-ood-diagnosis-derm1m · How It Works · Data refreshed daily, snapshot 2026-11-12.