Thanks for sharing your concerns, Philip. To properly investigate what's happening with the prompts, I can take a look at your specific table through its URL (just copy it from your browser's address bar). This will let me see exactly what's going on while keeping your data private.
While different AI models definitely handle prompts in their own unique ways and get regular updates, significant performance changes usually come down to the input data and prompt structure. Let me take a look at your specific case so we can see what might be causing these changes you're seeing.
Would you mind sharing that table URL so I can investigate further?