There is no single page count that applies to every ChatGPT conversation and API workflow. OpenAI’s general ChatGPT file-upload guidance states limits in file size and tokens—not pages: up to 512 MB per file and 2 million tokens per text or document file. These are upload caps, not a promise that every page will be included in the model’s active context or fully covered in an answer. (help.openai.com)
Whether a document works well depends on the product path, model, file contents, and task. A file may meet an upload cap yet still be a poor fit for a request that needs comprehensive coverage.
What the limits mean
| Limit or capability | What it tells you | What it does not tell you |
|---|---|---|
| ChatGPT upload caps: 512 MB per file; 2 million tokens per text or document file | Whether a file is within the stated general upload limits | A page maximum or a guarantee that every page will be covered |
| Enterprise context comparison: 128,000 tokens, described in OpenAI’s Enterprise guidance as roughly 200 pages of text | A rough comparison for the context window described in that guidance | A universal ChatGPT page limit or a guarantee that all uploaded text is placed in the prompt |
| API PDF input | How a specific API route accepts and processes a PDF | A page limit shared with ChatGPT uploads |
OpenAI’s Enterprise guidance says some relevant document text is included directly in context, with additional text accessed through search. It notes that this can work well for specific questions but may struggle with complex tasks such as summarizing very large documents or comparing multiple large files. So the guidance’s “roughly 200 pages” comparison is not a universal page limit or a measure of how much of every upload is read at once. (help.openai.com)
ChatGPT uploads and production APIs are different
For ChatGPT, use the upload limits above as file-acceptance checks—not as a page count or a guarantee that an answer reflects every page.
For a production workflow, OpenAI documents PDF input through the Responses API as an input_file. Its file-input guide describes base64 data or an uploaded file ID as input methods. On vision-capable models, PDF processing includes extracted text and page images. OpenAI recommends File Search for retrieval over large files rather than passing them directly as input_file. (developers.openai.com)
A concrete supported-path check:
- Model and context: GPT-6 Astra is listed as supporting the Responses API and image input. Its model page gives a 1,050,000-token context window, a 922,000-token maximum input, and 128,000 maximum output tokens. These are token limits, not page limits. (
developers.openai.com) - Count the actual PDF request: PDF input to a vision-capable model includes extracted text and page images, so page count or a plain-text estimate alone is not enough. The Responses API input-token-count endpoint accepts the same input format as Responses requests, including files and images. Use it to count the actual request, then compare the count with the model’s input limit and leave room for the response within its context and output limits. (
developers.openai.com) - Keep storage and direct input limits separate: The Files API’s 512 MB per-file limit applies to file uploads. It does not establish that a PDF of that size is eligible for, or suitable for, direct model input. The file-input guidance retrieved here does not specify a separate direct-PDF per-file or combined-request size limit; check the current file-input requirements for your chosen route rather than treating the storage cap as approval for a request. (
developers.openai.com) - Consider retrieval for large files: File Search is a different workflow for retrieving relevant information from uploaded files. It is not the same as placing the PDF directly in a request and does not guarantee exhaustive coverage of every page. (
developers.openai.com)
Why page count alone is a poor guide
Page count does not capture how much text or visual information a model must process. In the documented Responses API path, PDFs sent to vision-capable models are processed using extracted text and page images. The file-input guide also recommends File Search for retrieval over large files. (developers.openai.com)
For example, two 120-page PDFs may differ substantially: one may contain searchable text, while another is dense with charts and small print. Their page counts alone do not establish whether either fits a particular request or whether the answer will address every page. Check the limits for your chosen product path and model, then test the actual document with representative questions. If broad coverage matters, compare answers with the source and ask targeted questions about sections or page ranges. That is a practical review step, not a published page limit or guarantee.
Bottom line
ChatGPT’s general upload caps are 512 MB per file and 2 million tokens per text or document file, not a fixed number of pages. Separately, OpenAI’s Enterprise guidance compares a 128,000-token context window to roughly 200 pages of text; treat that as a guidance-specific comparison, not a universal product maximum. For a production API, GPT-6 Astra is one documented Responses API route: check the actual request’s token count against its model limits, and distinguish Files API storage from direct-request eligibility. Page count by itself cannot establish capacity or full-document coverage. (help.openai.com)
Sources
Optimizing file uploads in ChatGPT Enterprise