The 4k token limit refers to the output token limit which is the same across all of the latest models.
The 128k, on the other hand, refers to the total token limit (or context window), which is shared between the input and output tokens. Both gpt-4-turbo models and gpt-4o have a 128k limit/context window while the original gpt-4 has an 8k token limit.
I hope that answers your question.
Thanks for clarifying. I see in the screenshot below that my maximum context window is only 4096 tokens, not 128k. I believe this might depend on the account type. Do you know how to upgrade to the 128k token version? Perhaps contacting OpenAI support would be helpful?
As said, maximum tokens here refers to the output tokens. This is the same across all of the latest models as also shown in the model overview here.
To give you an example: If you are trying to get the model to produce an output of 1,000 tokens, you could include up to 127k input tokens as part of your system and user messages. So the sum of the input and output tokens cannot exceed 128k. However, the output tokens - at this point in the model evolution - is always limited to 4k.
Bottom line: There is no upgrade and there is no need to reach out to OpenAI support.
The screenshot I showed you highlights the ātotal context windowā setting, not the output.
For instance, my other model, the 3.5 Turbo 16k, has a token limit of 16,000, the screenshot below shows the context window setting.
I used Poe, which offers both GPT-4o (with a 4,000 token context window) and GPT-4o-128k (with a 128,000 token context window).
I canāt speak specifically to the Poe account you are referencing - however, the information / advice provided to you remains unchanged.
The discrepancies in terms of token limit you are seeing in the Playground is likely a bug. You can rely on the information in the model overview as per the link I shared which specify all relevant details by individual model.
I was looking forward for this answer for SEVERAL hours. Tyvm. Do you know if ScholarGPT ( ChatGPT - Scholar GPT ) uses GPT-4 or GPT-4o? I really need this answered so I can move on to my research project.
I need to know ScholarGPT Context window and GPT-4o context window (and if possible, GPT-4o mini).
Thanks a bunch!
All GPTās now use GPT-4o.
Iād love to read more about it. Do you have a source? Does this mean I have a 128k context window with all custom GPTs now? Thatād be awesome.
Also, Iām curious if we still have a limit on messages with GPT-4 and GPT-4o, like before (it was like 40 every 3 hours or something).
Thanks in advance 
No ChatGPT is manually limited to 16k.
Yeah, but it is now 80 messages
I guess myself.
Thatās great news, actually! I was considering whether to use Scholar GPT in the web for a research project. 16k token cap works for me. This will actually make my job like 10 times easier, no joke.
Tyvm, sir!
Iām confused what you mean by āmanuallyā (sorry Iām just starting to learn about AI), is it the normal limit on the web browser without API or anything similar?
What does it mean by āOutput is always limited to 4kā?
can we not change it? I have paid ChatGPT subscription.
I have no idea what language youāre speaking in. Is there any place that talks in laymen terms what ā128k context lengthsā mean in english? Does it have to do with how many characters I can use in my questions?
