An iPhone app called Soba, designed to estimate meal carbohydrate content for type 1 diabetes management, was benchmarked against eight AI models using 400 meal photos. The testing methodology involved controlled datasets and a holdout set to ensure accuracy. Gemini 3.7 Flash initially showed promise, outperforming the app's current model, but a tie on the holdout set and higher costs led to the continued use of the existing Gemini 3.6 Flash. AI
IMPACT This detailed benchmark provides insights into the practical performance of various LLMs for specialized tasks, informing developers on model selection for niche applications.
RANK_REASON The article details the benchmarking of AI models for a specific application (a carb-counting app), rather than a new model release or significant industry trend.
- ChatGPT
- Claude Sonnet 5.5
- Gemini 3.6 Flash
- Gemini 3.7 Flash
- iPhone
- Nutrition5k
- SNAPMe
- Soba
- type-1 diabetes
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →