So far Opus 5.5 is great for me. I did read this guy's substack below regarding Opus 4.6 and to be honest I never used 4.6 because it was already an older model when I started using Claude. I thought the article was interesting, so I tested it on a small writing project that I had and the 4.6 result was the best of the bunch. Don't know if that would persist across different domains, but the result for this particular task was more human than the other models. https://www.whytryai.com/p/opus-5-5-vs-opus-4-6-vibe-check