GPT-5 Series for Dermoscopic Image Labeling

The launch of ChatGPT has raised interest in Generative Artificial Intelligence models and the way they can be applied across domains. In this study we focused on dermoscopic image labeling, using OpenAI's GPT-5 series to determine if there are improvements in labeling between the last versions: GPT-5, GPT-5.2 and GPT-5.4. The testing process uses three different formats: free answer, choosing between labels, and choosing between labels after providing patient metadata together with the images. The evaluation was carried out with paid ChatGPT Plus subscription deactivating the option "Improve the model for everyone" ensuring an unbiased testing. The models were tested using the Thinking variant during different timeframes. The results underline improvements in distinguishing melanoma across model iterations.