In my experience, ChatGPT (the free tier) tends to use the dataset that it already has, even when it has the ability to access the internet. So when i use it, i always specifically “push it” to go online and find the most up to date information. And it actually does. But of course that is an unnecessary hassle for me to include that in the prompt.
Grok on the other hand, when i ask something, goes directly to the internet. I can actually see it, it tells you what it is doing while processing. However, that doesn’t mean what Grok gives me is the most recent, i also ask it to verify, but in my experience, Grok is better in terms of recency.
I always prefer the models with full internet access. I remember the times that the answers were a couple years old with dataset cutoffs (it wasnt too long ago). AI was only useful like an “encyclopedia” back then, hopefuly young people will know what i am talking about 
I see that there are a lot of models out there now. Maybe we need something like perplexity to at least moderate or approximate the outcome that we can receive from multiple models to save time. It is funny that now we need another AI to regulate the so many different AI responses. AI for AI.