A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosis

A 10-year-old convolutional neural network (CNN) named Inception v3, fine-tuned on public dermatology images, outperformed two frontier language models (LLMs), GPT-6 Astra and Claude Fable 5.1, in diagnosing skin cancer in 2026. The CNN correctly identified 62% of 200 biopsy-proven lesions based on dermoscopic images, while the LLMs achieved 42% accuracy.

The CNN was also better at diagnosing harmless lesions, such as moles and seborrheic keratoses, compared to the LLMs. Both the CNN and LLMs received the same prompt and had access to the same images, but the CNN achieved higher accuracy scores. The CNN's performance was consistent across different diagnoses, including melanomas, seborrheic keratoses, and basal cell carcinomas.

The CNN's ability to perform well on these tasks demonstrates its potential as a valuable tool in dermatology, particularly for diagnosing benign skin conditions.