Opus 4.8 to Opus 5.5
retained text
I personally tried all new models but I always go back because it took me a lot of time and experiments to fine tune current rules, skills and flows. I would need to redesign all my skills probably in order to have the same performance, but this feels rather time consuming because I would need to first try it, then audit it, then adjust. This would take weeks. I am also not sure why Opus 5.5 is being described as really great and efficient because in my results it reads way too much, doesn’t think about the given task too much nor does it respect the rules. In the last audit it simply said that there was a (let’s call it) shortcut in the skill and it went that way. The claude code development experience does seem better and better, but the models most of the time seem off. How is your experience so far?
document fetched 2026-09-26 00:40 · canonical https://reddit.com/r/ClaudeCode/comments/1wpydgg/opus_48_to_opus_55