Interesting. It’ll probably be quite a step up from GPT-4o, which is very much behind the SOTA by now. What do you mean by graphics coding? Feeding images of the code? Or screenshots of the state of the UI?
Discussion
I am not a coder. I tried I guess what is being called vibe coding by having a conversation with ChatGPT. While it was very quick to spit out code that I could paste into Swift Playground, the whole experience started to derail when I started to change my mind about what I asked for. To me it seemed logical that I needed to start with a macro view. Because I think in pictures easily I decided to create a document with screen mockups accompanied by annotations describing functions. My idea is to upload the doc and ask for code based on the macro package. Once the doc is complete I thought I would compare 4o and Sonnet. Probably a dumb idea but what do I know?
Not completely sure I understand what you're doing, but from my experience older models like 4o really struggle with programming languages like Swift that aren't as popular. Uploading images of UI mockups and a general idea of what you want the LLM to do is probably a viable strategy. The biggest thing you have to be cognizant of when you're coding using a chat interface is that you keep the file context updated for the model.