It's substantially better.

I've compared the same prompts between both, and GPT-4 is always better, except when you need speed or it's a very simple task.

For complex tasks, especially requiring some sort or social understanding (I'm building a group AI sharing tool) it has a capacity for nuance which is sometimes surprisingly good. ChatGPT is usually more straightforward.

If anyone hasn't played with gpt4 yet I'm looking for testers on my platform and I have access.

It’s far better at producing workable code, logical reasoning, keeping a train of thought, etc. Their benchmark breaks it down.

https://openai.com/research/gpt-4

It is much better, yes. The new 3.5 is still pretty good and very cheap, so it's my go-to for simple things, but GPT-4 is unmatched for complicated reasoning.