Opus 4.1 is by far the best right now for most tasks. It’s the first model I think will almost always pump out “good code”. I do always plan first as a separate step, and I always ask it for plans or alternatives first and always remind it to keep things simple and follow existing code patterns. Sometimes I just ask it to double check before I look at it and it makes good tweaks. This works pretty well for me.
For me, I found Sonnet 3.5 to be a clear step up in coding, I thought 3.7 was worse, 2.5 pro equivalent, and 4 sonnet equal maybe tiny better than 3.5. Opus 4.1 is the first one to me that feels like a solid step up over sonnet 3.5. This of course required me to jump to Claude code max plan, but first model to be worth that (wouldn’t pay that much for just sonnet).