I personally prefer the ‘tone’ of gpt-4o.
Occasionally when I get stuck with something using gpt-4o, I try claude-3.5-sonnet, because I have heard online that it is better for coding tasks etc, but, for me, it often doesn’t seem to improve the situation, and I find the responses are a bit more ‘dry’ (others may prefer that though).
It’s still great to have all the different models available and try one when another is not producing the results you want. And I think they are all in a state of swift change, so they could all get much better very quickly.
In Cursor, I also switch to gpt-4o-mini for smaller tasks and questions as it is not considered a premium model and therefore doesn’t count towards the 500/month Pro user premium quota.
I should add, most of the time the ‘stuckness’ I experience comes from trying to achieve non-standard implementations of things or using solutions or SDK’s that are not well documented.
For more standard solutions, or learning the basics, I find all of the models are pretty helpful.
It does help if the prompts you provide are clearly thought out and precise - trying to get something new working when you are tired and not thinking clearly often doesn’t end well for me.
For reference, I found the video below had some good ideas about how to use AI in software development, such as having reasonable expectations, developing iteratively and focusing on one problem at a time. The presenter mainly uses an Aider and Claude stack, but I think the principals apply well to all AI stacks.
Mastering AI for Software Development: Essential Tips & Principles
(13 mins)
by Coding The Future with AI
https://youtu.be/VrHXjvikBrY
Good luck!