| Mobile Web

Older Google model beats latest GPT and Claude in skin cancer test

Latest general-purpose AI models recorded lower accuracy than Google’s Inception v3 image recognition model released in 2016 in a skin lesion diagnosis experiment. Researchers compared GPT-6 Astra, Claude Fable 5.1 and Inception v3 using 200 lesion images split between cases needing treatment and those requiring no action. Inception v3 led overall and in melanoma identification. The study said differing training conditions limited interpretation and larger validation and safety assessments are needed for clinical use.