Estonia's Haut.AI scores 69.6 as rival AI models flag skin problems that aren't there
What's the deal? Estonian startup Haut.AIDealroom has a profile for this one. Try Dealroom → has launched SkinFM, a foundation model for skin analysis, alongside SkinBench, a public benchmark that pits its technology against six general-purpose and medical AI models. Its own tests found rival models frequently flagged skin concerns that experts said were absent.
By the numbers: Haut.AI scored 69.6 out of 100 on SkinBench's Skin Intelligence Score. Google's Gemini 3.5 Flash led the general-purpose field with 50.6, followed by OpenAI's GPT-5.6 Sol at 45.9 and AnthropicDealroom has a profile for this one. Try Dealroom →'s Claude Fable 5 at 45.4. Models from xAI and Alibaba and Google's medical MedGemma 27B trailed.
The findings: Across 19 tests covering wrinkles, pigmentation, redness, dark circles and eye puffiness, the six rivals reported eye puffiness in roughly 95% to 100% of images where none existed. They also struggled to locate wrinkles accurately — one model marked the nose and mouth instead of frown lines.
What it does: Haut.AI, founded in 2018 by Anastasia Georgievskaya and Konstantin Kiselev, sells skin-analysis tools to beauty brands and retailers, including Ulta and BeiersdorfDealroom has a profile for this one. Try Dealroom →. SkinFM was trained on more than four million photos of real skin and lets brands analyse multiple skin traits from one model rather than building a separate model per task.
What could go wrong? Haut.AI designed and ran the benchmark itself while selling competing technology. Some tests also used computer-generated faces — 350 synthetic images were used to assess glabellar wrinkle severity and location.
Context: Consumers are already leaning on AI for skincare. A Haut.AI survey of 1,238 US women found 62% use or are interested in AI for skin or body advice, and NielsenIQ found 49% of consumers have acted on beauty recommendations from generative AI.
The signal: The launch tests a claim central to the beauty-tech market: that a model able to describe a photo can also measure skin. Gartner predicted that by 2027 organisations would use smaller, specialised models at least three times as often as general-purpose ones — a bet Haut.AI is now pressing with its own scoreboard.
Read more: geeniusdigipro.delfi.ee
Image credit: Haut.AI
