As we use the AI tool I assume that it also helps train that LLM to be better.
Is there a particular one that perhaps we should focus on using or what one do people use for themselves?
Claude3.7 is the one I tend to use
I highly doubt Wappler AI interactions will cause a model to be better. Usually it's trained on public code, and in the cases where the AI provider does save interactions to enhance the model, it's unlikely anyone working in AI supervision will know anything about Wappler 
Claude 3.5 and 3.7 seem to be the leaders in Agent mode, but 3.7 has a tendency to make less desirable workarounds when it can't do something (e.g. a variable isn't being saved properly? ok, let's remove it completely)
Claude 3.5 and 3.5 Sonnet perform indeed best, but also the new GPT 4.1 is very good and also very fast