I used to trust OpenAI models more for backend work.

Claude was my go-to for frontend.

Opus 5.5 completely flipped that.

Claude already nailed frontend, but now it’s seriously fucking good at backend work too. I’m using Opus across the stack, and Codex has basically become my second reviewer.

That’s a pretty wild reversal of my workflow.

Whatever Anthropic is doing with post-training deserves to be studied. So does how OpenAI went from my default for backend work to a second opinion.