Opus 5 was just incoherent - curious to see what improvements they have made here. Would love to see some kind of postmortem to better understand how writing styles change from model to model.

I wouldn’t be surprised if Opus 5 was trained on content written by other LLMs