Haut.AI said its new SkinBench comparison found that six other AI models often flagged images as showing eye puffiness when it was absent, a potential concern for beauty brands that use automated skin assessments.
In its Oct. 27 announcement, Haut.AI reported false detections in 95% to 100% of images without eye puffiness. Its SkinFM system had an approximately 18% false-detection rate on the same task.
The 19-test comparison covered seven skin features: eye puffiness, dark circles, skin type, skin tone, redness, sun spots and wrinkles. Eleven tests asked models to identify or grade features; the other eight assessed how well they located them. Haut.AI said it compared the results with expert assessments and defined a score of 100 as perfect agreement with experts on every image.
Haut.AI reported that SkinFM scored 69.6 out of 100 overall. Gemini 3.5 Flash, the highest-scoring general-purpose model, scored 50.6. The other models in the comparison were GPT-5.6 Sol, Claude Fable 5, Grok 4.6, Qwen 3.8 27B and MedGemma 27B.
Results varied by assessment. Haut.AI said SkinFM detected 82% of deep forehead wrinkles, while the other models ranged from 15% to 60%. It reported a score of 77.8 for SkinFM on skin-type classification, compared with 57.7 for its strongest competitor, but said Gemini did better on crow’s-feet grading and skin-tone assessment. Some frown-line tests used 350 computer-generated faces; the announcement did not give sample sizes for every test.
Haut.AI provides skin-analysis software to beauty brands, where selfie assessments can inform product suggestions. The company said inaccurate readings could lead to unnecessary recommendations and spending and reduce consumer trust. Its announcement did not report testing purchases or whether recommended products worked.
Haut.AI said SkinFM’s training set contained more than 4 million images of real skin, and that brands can use the model to develop tailored measurements and grading systems. “AI can recognize a face, but that doesn't mean it can accurately assess the skin,” said Anastasia Georgievskaya, Haut.AI CEO and co-founder. The company plans to add more models and consistency tests to SkinBench, but has not given a timetable.
