I got early access to ChatGPT API – here’s what you need to know (buildt.ai)

You probably aren't allowed to disclose that secret sauce as it is different from the publically-released API.

> I’ve noticed its likelihood to adhere to its instructions (particularly if they’re complex and logic based) the closer you get to the max 4k tokens context window limit.

I had heard ChatGPT had a 8k token context window, is that information wrong or is the API one less capable?