Having trouble accessing GPT-4? Here's what you need to know

If you have been granted GPT-4 access (you would have received an email), it is only granted to the organization specified in the waitlist form that you applied with, access cannot be transferred to another account. You can specify your organization in your API requests with the header:


Additionally GPT-4 models are only supported through the v1/chat/completions endpoint and will not work with v1/completions. You can read more in our Chat Completions guide.

You can try our GPT-4 in the playground as well if you have been granted access. Make sure to set the playground to “Chat Mode”.


but why it always show “I am a part of OpenAI’s GPT-3 model, which is a powerful natural language processing model capable of answering questions, writing articles, and performing various tasks. GPT-3 is the latest released version as of now.” i am sure i used gpt-4 and added “OpenAI-Organization”

I requested GPT-4 API access on March 14 and have heard nothing as of April 5. At what point should I start to worry that we’re being singled out or that our waitlist application was misplaced?

It seems to me that if it’s possible to give every Plus subscriber limited access to ChatGPT-4, it should be equally possible to give every API customer limited API/Playground access to the GPT-4 model. Why not use the waitlist for rescinding limits instead of blocking us from doing any experimentation at all?

Hey Logan,
Thank you for sharing with us. However I have received the GPT-4 API access invitation email on March 28, but I still cannot find GPT-4 in my model list and cannot find the access to GPT-4 in the play ground either (both for ChatCompletion). Could you please help me to check my account why it does not have the gpt-4 api access?

1 Like

Hey @hemp there are still a lot of folks who don’t have access yet, though this will change at the end of the month! Currently, all developers who made a successfully payment on their account of $1 or more have access.

@xingyudang looks like this is resolved!