Accuracy of ChatGPT , Gemini, Claude and DeepSeek in Carbohydrate Counting

Key Points

• The research aims to assess the accuracy of various AI models in carbohydrate counting compared to clinician calculations.
• Evaluated the accuracy of ChatGPT, Gemini, Claude, and DeepSeek in carbohydrate calculations.
• Compared AI outputs against clinician standard measurements.
• Applied clinical equivalence criteria to assess AI performance.
• ChatGPT provided the most accurate carbohydrate estimations among the tested models.
• Other models underestimated carbohydrate content, but deviations were clinically acceptable.
• Findings support the use of ChatGPT as an effective tool for individuals with type 1 diabetes.

Abstract