Did Opus 5.5 change under the hood on November 3?
Shane Melaugh had Gemini 4 Argon analyze 1,400 of his Claude Code sessions and says it found a sharp shift in how Opus 5.5 reasons starting November 3, with no change to the model name.
I've been super happy with Opus 5.5. I think it's one of the best models released in terms of capabilities, though still too expensive for API use, in my opinion. So if this turns out to be true, it would be very disappointing.
It's one tweet and one person's sessions, so I'm treating it as a question worth watching, not a conclusion. But the method is interesting: using one model to audit another across a big pile of real work logs is a practical way to notice drift you might otherwise just chalk up to a bad day.