Quesiton About "8K" GPT-4 Model

I received the email saying “You can now access GPT-4 models with 8K context via the existing OpenAI API.”

Is this supposed to apply to the playground as well? In there, when I Select: Mode=ChatGPT > Model=GPT-4, the “maximum length” slider only goes up to 2048. And GPT-4 is not a model option for “completion” mode.

Also as I read the documentation page, for “Model endpoint compatibility”, I do not see anything mentioning GPT-4 8K, only 32K or regular “gpt-4”. I assume the regular “gpt-4” the 8K one based on the 8192 token limit in the table at the top. But was confused by the lower limit in the playground.


Yeah, in the Playground for me, just got the invite

Hard refresh your browser, it’s under “Chat”


Right, I see that too. I should have clarified better, but my main question is whether the playground GPT-4 is supposed to have the 8K token limit. Because right now it appears to only go up to 2048 when dragging the maximum length slider.

My slider has max 2048 too. Maybe the UI in the Playground is glitched. Have you tried 8k in the API yet?

1 Like

It could be that it’s simply not available for playground.
Just as temperature is locked to 1 instead of 2.

Actually, the HTML indicates (and for me, can) that it goes up to 4,000

<div tabindex="0" class="rc-slider-handle" role="slider" aria-valuemin="1" aria-valuemax="4000" aria-valuenow="4000" aria-disabled="false" style="left: 100%; right: auto; transform: translateX(-50%);"></div>

Perhaps a CTRL+R is in order?


To be hair, things are moving at lightning speed… hard to keep up with docs and little details…

(ETA: Leaving “to be hair” in to prove I’m a human writer? haha…)

The API is coming back slow for GPT-4 too, so I’m not pounding it yet. Gonna get some stuff done around the house then play more later tonight, I think.

It’ll come. Patience is a virtue! :wink:

1 Like

Yes, I noticed the same. I hope they allow the playground to use the max tokens. The playground is a much cleaner interface than my Python code!

Yeah, I’m having the same issue. In the playground I can only see GPT4 in Chat mode, not in Completion mode.

right now, API is so slow. It is nowhere comparable to 3.5. It is a little disappointment with the speed. Do we know that it will be fixed soon?