I've noticed the same! I assumed it was something to do with its reinforcement learning making it more organic and less LLM-speak, but that sounds plausible.