This is contextual, the model seems to match the user vibe and the general situation. Take a look at the "Customer service proof of concept" video added to the announcement post[1] - there are two AIs talking using very different voices.
[1] https://openai.com/index/hello-gpt-4o/