Sonnet still better than o3-mini?

Discussion
1

Hey there,

I just tested o3-mini on a fairly simple error I was getting, and after three attempts where I would just introduce new bugs, I gave up and switched back to Sonnet, which got it one go, changing only three lines of code. Am I the only one experiencing this?

I have had this happen a few times now. A new model releases that scores better than Sonnet. I test it, and it’s okay, but never as good as Sonnet, so after a bit, I go back to Sonnet. It might just be me, being much better at prompting Sonnet or something like that.

Also, not having the agent mode for the other models is a big downside for me.

2

I use Sonett; if it fails, try o3, DeepSeek, and Gemini. in most cases start with sonnet. It has correct tool usage in agent mode. In chat, o3 may be ok too. Open AI models worked better on larger codebases if you needed to provide more context than Claude could handle. Now with agents, it’s a bit less of an issue.

3

its kinda simple.
There is not a single LLM which has better Frontend code than Sonnet. For backend and complex task i recommend r1,o3

2 Likes