How to get access to gpt-4-32k?

Hi all. How does one go about getting access to the 32k models? Our team has been approved for the 8k models, but our use cases require the 32k models. Any assistance would be appreciated!


I saw this a few days ago. Maybe this will help you: GitHub - openai/evals: Evals is a framework for evaluating OpenAI models and an open-source registry of benchmarks.

I’m having the same problem. Will I be able to access the 32 k version in the playground?

Yes, when it is available.

32k is not yet available.


1 Like

Is there any ETA that you know of? How long usually does it take before they release the full product from what seems to be in this case like an advanced beta?



OpenAI is experiencing a lot of growing pains.

Sorry to be so honest.

Its will be released when it is released.


1 Like

Alright, we’ll just have to wait then, thanks!


According to this, it should be available? :thinking: I do have access to GPT-4 but not 32k


Looks like the 32k models are being rolled out separately:

If you want an interactive CLI to the API (similar to ChatGPT), feel free to check out this project I created (supports all models):

Looking forward to using the 32K model. Any ETA?


Is there any tutorial on the evals on how to install and use it? Would be greatly appreciated. Although I’m new to programming I wish to learn as much as possible by learning as I go.

Looking forward to this as well… :star_struck:
Hopefully, soon in my playground!

Greetings from Berlin

1 Like

What I do for my startup is I have a for loop run through the text with 3.5 and summarize/compress the text. I end up getting the 14,000 token transcript down to about 1,000 tokens and it works great.


Yeah it seems weird. I expected to have access based on the docs and such.

Calling the available models endpoint seems to give a true list of what you have access to (I haven’t found much other use for it).

Does anyone know if there’s a way to get notified when you have access to new models? For now I will just call that endpoint frequently, which seems laborious and wasteful, and I still might go days with access before I realize I have it.

1 Like

I have the same question. It’s painful seeing people with early access to multiple features you would love to start building with.

1 Like

I was aware of getting 32k access by randomly looking at the models under Chat in the Playground.

But a cron job that periodically sends data to the 32k endpoint, and sends you an SMS or email when it it successful is probably the best you can do on your own right now.

1 Like

Why not just query instead?

1 Like

This probably makes more sense than my original suggestion since it doesn’t involve cost, so yah, good catch.

Didn’t you get an email when you got access?

I love your thought, I had a script running that turned on a lamp when I got GPT-4 access. Ironically I turned it off about a week before I got access :rofl:

No, no email. Just randomly dropped into the Playground, clicked Chat, and saw the 32k model in the dropdown. I could have had it days earlier for all I know.

I dropped what I was working on, tried it in the API, and it worked!

Now just trying to wrap up GPT-4 integration before I can get serious with 32k. But I probably need to up my quota from $120 per month, to something like $1000 if I am going to use 32k and normal 8k GPT-4 at a larger scale.

So now, just feeling out the cost and performance of GPT-4 before dabbling with 32k.

OpenAI is moving really fast these days, it’s hard to keep up!