I dunno, my experience mirrors the parent posters: we use opus for all our coding, but gpt 5.4 for all of our enterprise agentic work via api (much bigger amount of tokens). it just seems to be more optimized for this.

It’s much cheaper, for starters.