Hey folks! I've been using Claude 3.7 Sonnet for my coding projects and have now switched to Claude Code thanks to my MAX subscription. However, I've noticed it tends to complicate things unnecessarily and often ignores my clear requests to keep it simple, following principles like KISS, DRY, and YAGNI as outlined in my CLAUDE.md and instructions. I sometimes forget that Opus exists, so I'm curious—how do Opus and Sonnet 3.7 compare for coding and math tasks? I'd love any insights! Thanks!
5 Answers
I feel you! I’m in the same boat here, waiting to hear what others say. I’m currently using Gemini 2.5 Pro and GPT-4.1, since Sonnet 3.7 is a bit pricey to juggle around.
I had a rough day with 3.7; it kept hardcoding things despite my requests to avoid that. I even had to switch to Opus because it just wouldn't listen. It’s disappointing because when it works, it’s great, but it can get stuck on ideas and ignore your guidance.
I’m in the same boat, feeling like Claude has become unusable lately. I’m even considering switching to 3.5 since it seemed to have a better focus overall.
Don't forget about the context window. Sonnet 3.7 can struggle with longer contexts, similar to other models. Also, it’s key to give specific examples when you tell it not to over-engineer. Just saying 'don’t over-engineer' might not mean the same thing to it as it does to you.
I’ve recently started using Opus for my coding, and I’ve seen an improvement. It requires fewer corrections from me, though it’s still early to give a full assessment for it.
Using adversarial AI has helped me in cases like this. I upload my code to Gemini and ask it to critique it where I didn’t follow the required methods, then it gives me better suggestions.