# SkinBench

> SkinBench is Haut.AI's benchmark for AI skin measurement. It tests how accurately AI models measure skin: Haut.AI and six other AI models are scored on the same images against expert ground truth, using identical test sets and published metric definitions. Published 2026-10-27.

Scope: 7 skin parameters (wrinkles, sun spots, redness, skin tone, skin type, dark circles, eye puffiness), 19 tasks (11 grading, 8 tracing), 7 AI models.

## Leaderboard: Skin Intelligence Score (out of 100, higher is better)

The Skin Intelligence Score is the average of the seven skin parameter scores.

1. Haut.AI (Haut.AI): 69.6
2. Gemini 3.5 Flash (Google): 50.6
3. GPT-5.6 Sol (OpenAI): 45.9
4. Claude Fable 5 (Anthropic): 45.4
5. Qwen3.8 27B (Alibaba): 41.7
6. Grok 4.6 (xAI): 39.2
7. MedGemma 27B (Google): 28.5

## Scores by skin parameter (out of 100)

Model | Wrinkles | Sun spots | Redness | Skin tone | Skin type | Dark circles | Eye puffiness
Haut.AI | 61.5 | 52.6 | 91.5 | 87.6 | 77.8 | 52.7 | 63.5
Gemini 3.5 Flash | 39.2 | 28.3 | 89.6 | 90.4 | 43.4 | 32.4 | 31.2
GPT-5.6 Sol | 39.1 | 32.4 | 72.9 | 81.6 | 57.7 | 25.2 | 12.2
Claude Fable 5 | 37.6 | 34.0 | 71.5 | 88.7 | 41.0 | 21.2 | 23.8
Qwen3.8 27B | 31.5 | 25.9 | 77.0 | 86.3 | 47.0 | 7.4 | 17.0
Grok 4.6 | 31.3 | 26.7 | 61.4 | 87.9 | 36.8 | 13.6 | 17.0
MedGemma 27B | 14.9 | 20.6 | 60.3 | 54.4 | 36.6 | 4.1 | 8.6

Haut.AI leads on 17 of the 19 tasks. It is behind on skin tone and crow's feet grading.

## Sections

- [Leaderboard](https://skinbench.ai/#leaderboard): Skin Intelligence Score and summary scores for all seven models
- [Scores by skin parameter](https://skinbench.ai/#parameters): scores on each of the seven skin parameters
- [FAQ](https://skinbench.ai/#faq): method, datasets, statistical significance and update schedule
- [Contact](https://skinbench.ai/#contact): get in touch with Haut.AI

## About

SkinBench is published by Haut.AI (https://haut.ai), a B2B provider of AI-powered skin and hair analysis technology. Book a demo: https://form.typeform.com/to/xvdyKvB2