The jump in reasoning quality is noticeable. What's interesting is how it handles ambiguous instructions now — it seems to ask fewer clarifying questions and just makes a reasonable judgment call. That's a double-edged sword depending on your use case.