Shane Melaugh, a heavy Claude Code user, says Opus 5.5 changed how it reasons on 3 November, and that its “high” effort setting now thinks much less than it used to. He posted the claim as a thread on X, captured by Thread Reader App. The thread carries no response from Anthropic, so nothing in it confirms a change on the company’s side.
His evidence is an analysis of 1,400 of his own Claude Code sessions, done by Google’s Gemini 4 Argon, which pointed to 3 November as the day something shifted. The model’s name stayed the same. The amount of reasoning at “high” dropped sharply, he says.
He also ran a test on a real bug. At “high,” Opus 5.5 rushed in, got confused, and had no fix after three hours. At “xhigh,” it paused to think and found one within minutes. His conclusion is that the popular habit of using “high” for everything no longer holds.
One bug is an anecdote, and a second AI doing the reading adds its own room for error. Anyone running Opus 5.5 at “high” can settle it cheaply by timing it against “xhigh” on a task they already know well.
Reported by Thread Reader App from a thread by Shane Melaugh on X; the source carries no publication date.
