SkinBench shows general AI models flag eye puffiness in 95-100% of clear-skin images
Individual-War3274 · reddit · 2026-10-27
SkinBench, a new public benchmark, tested six general-purpose AI models (including GPT-5.6 Sol, Gemini 3.5 Flash, Claude Fable 5, and Grok 4.6) against a specialized skin-analysis model, and the accuracy gaps are stark:
- False positives: the six models identified eye puffiness in 95-100% of images where it wasn't present
- Wrinkle detection: the specialized system caught 82% of deep forehead wrinkles vs 15-60% for general models
- Skin-type classification: 77.8/100 for the specialist vs 57.7 for the strongest general model
It's not a total loss: Gemini did better on some tasks like crow's-feet grading and skin-tone assessment. The takeaway: recognizing what's in an image isn't the same as accurately measuring a specific characteristic — a real concern as general models become the default for consumer advice.
More from Models
- Microsoft's MAI Cyber model doubles user adoption to 50%, gains 20% token efficiency — mustafasuleyman · 2026-10-28
- Microsoft's MAI Cyber model doubles user adoption to 50% in 30 days — mustafasuleyman · 2026-10-28
- Google's Genie 4 simulates photorealistic worlds with native audio in real time — natanielruizg · 2026-10-28
- Leak: DeepMind's Genie 4 is a real-time world model with native audio and type-to-edit scenes — rohanpaul_ai · 2026-10-28
- NVIDIA Launches Post-Training Nemotron for Europe via Nemotron Labs — NVIDIAAI · 2026-10-28
- AI model prices today mirror Uber's subsidized early days, may rise sharply later — pstAsiatech · 2026-10-28