Main takeaway: it’s not really “worse”, it’s trade-offs.
Better at coding/benchmarks, but worse UX (token usage, less pushback).
The sycophancy point is real — a model that doesn’t challenge you can lead to bad decisions.
What I genuinely can't understand is how Anthropic missed the sycophancy. A huge part of what drew people from ChatGPT to Claude was exactly that, Claude would push back. That was the differentiator. Removing it feels like a company disconnected from why its customers chose it.