Just an update !
Got an answer from a human at OpenAI which clarified things a bit.
It seems most ChatGPT Plus Users have a 128k context window so far.
Difference with the API likely comes out from differences with the instruction (ChatGPT have custom instruction to act as a itself, consider memory, contextual memory from other chats and so on.)
Also, regarding differences in managing big documents, it’s my understanding that if you use the upload document using the upload file interface from ChatGPT, the whole document is not uploaded into the context window I think ; its like the file search API function, which chunk your document in possibly 800 tokens parts and then section linked semantically to the input are extracted and fed to the LLM for an answer… Thus the model do not see the whole document unles you copy the whole text directly in the context window (or canvas).
We understand that you came across the 32k token limit on our website, and we truly appreciate your feedback regarding our product information. Your comments have been noted and will be reviewed internally.
We’d also like to clarify the earlier mention of a 200k token limit — that specific limit applies only to our Enterprise and Education offerings. You can find more details here.
For ChatGPT Plus users, models such as GPT-4o and GPT-4.1 mini support a 128k token limit, which is significantly higher than the previously mentioned 32k. For additional information on token limits and model capabilities, please refer to this page.
Thank you again for bringing this to our attention. If you have any further questions, we’re happy to help.