A PDF is "too large" or only partly read because of one of five limits: file size, page count, text length in tokens, number of files, or how often you can upload. As of October 2026, ChatGPT accepts documents up to 512 MB and 2 million tokens each, Claude takes 500 MB and 1,000 pages per PDF in a chat, and NotebookLM takes 500,000 words or 200 MB per source. For text documents, the page and token limits are usually the ones that matter, not megabytes.

Every number below comes from the vendor's own help or documentation page, checked on 27 October 2026, with the link beside it. Where a vendor publishes no number, the table says so.

A tall stack of paper beside a desk tray holding a smaller clipped stack.
A document can fit in megabytes and still be too long for the tool.

The limits side by side

Limits as of October 2026. Vendors change them, so open the linked page if the number matters for your task.

LimitChatGPTClaude (chat)NotebookLM
File size512 MB per document or presentation; about 50 MB for spreadsheets and CSV; 20 MB per image500 MB per file; 30 MB per file in a Project200 MB per uploaded file
PagesNo page limit stated1,000 pages per PDF; visuals analyzed only up to 100 pagesNo PDF page limit stated; Google Slides up to 100 slides
Text length2 million tokens per text or document file (not applied to spreadsheets)Everything must fit the model's context window: 200K, 500K or 1M tokens depending on the model500,000 words per source
Number of files5, 25 or 40 files per project depending on plan; 10 knowledge files per GPT20 files per chat; unlimited in a Project if the content fits the context window50 sources on the free plan
How often80 files every 3 hours; free users 3 files per day; may be lower at peak hoursUsage limits that reset every five hours, plus weekly limits; no file counts publishedCompute-based usage limits; quota refreshes every 5 hours up to a weekly limit

Sources: OpenAI, Uploading files and audio to ChatGPT; Claude, Upload files to Claude and context window on paid plans; Google, Add or discover new sources and usage limits. Google's help center now titles NotebookLM "Gemini Notebook".

You may see "Claude takes 30 MB" quoted. On Claude's page, 30 MB is the per-file limit for Project files. A file attached to a chat can be up to 500 MB.

What the vendors do not publish

  • OpenAI's upload page gives no page limit for a PDF and does not state the context window of the model you are chatting with. The 2 million token cap is per file.
  • Google's NotebookLM pages give 50 sources for free users and no source count for paid plans, and they describe usage limits only as "standard" or a multiple of it.
  • Claude's help pages publish no number of messages or files you can send before a usage limit applies.

Figures for these on other sites are not from the vendors, so this article leaves them out.

Which limit did you hit?

What you seeWhat the vendor's page saysWhat to do
Claude: "Uploaded file is too large"You cannot upload a PDF over 1,000 pages, and this is the error shown.Split the document by chapter, or upload the text of the parts you need.
Claude reads the text but ignores chartsPDFs of 101 to 1,000 pages are processed as text only.Keep charts you need answers about in a PDF of 100 pages or fewer.
Claude: the chat hits a length limitLength limits come from the context window, which depends on the model.Start a new chat, or put the files in a Project.
ChatGPT: "upload limit reached" after a few uploads80 files per 3 hours, 3 per day on free; failed uploads can count toward the limit.Wait, check Settings > Storage, and send fewer, larger files.
ChatGPT misses material from a long fileFiles are capped at 2 million tokens. Outside Enterprise, ChatGPT uses text-based retrieval that extracts digital text rather than embedded images.Ask for the page or section behind each answer and check it against the original.
NotebookLM rejects a sourceLimits are 500,000 words or 200 MB per source, and 50 sources on the free plan. Sources that became inaccessible still count toward the limit.Split the source, or remove sources you no longer use.

Why a small PDF still fails

Megabytes measure how the file is stored. The limits that decide whether a tool can read it count pages and tokens.

Claude's PDF documentation says each page is processed both as extracted text and as an image. Text typically costs 1,500 to 3,000 tokens per page depending on density, and the image is counted on top of that. The same page notes that dense PDFs with small type, complex tables or heavy graphics can fill the context window before the page limit is reached. These figures are from the developer documentation for the API; the chat limits are on the pages linked above.

Claude's help page also says that for documents other than PDF it extracts text only and cannot read embedded images. A text file has no page images to process.

Scanned PDFs are a separate case. A scanned page is a picture of text, so there is no digital text to extract until OCR has read it. The vendor pages linked here do not describe how scanned pages are handled, so test a file by asking the tool to quote a sentence from a scanned page.

Convert to Markdown before you upload

Markdown is plain text, so the tool receives words instead of page images. MD Factory converts PDF, Word and other documents to Markdown on your PC, with no upload. Scanned pages are read with local OCR.

  1. Add the PDF to MD Factory, or drag it onto the window. Conversion starts right away.
  2. Wait for the row to show Completed, Needs review or Failed. Open Needs review results and compare them with the original.
  3. Open the .md file saved next to the PDF. Remove sections you do not need, such as a table of contents or appendices, and upload the result.

The original PDF is not changed. Pictures go into a _images folder beside the Markdown file and are not part of the text you upload.

ChatGPT's and Claude's lists of supported file types do not include .md; NotebookLM's does. If a tool does not accept the file, paste the text into the chat, or save a copy as .txt, which both lists include.

MD Factory does not split files, and it does not promise a smaller result: how long the Markdown text is depends on the document. Convert, upload, and see whether the tool now reads the whole file.

For a scanned PDF, see the scanned PDF guide. For a PDF that has selectable text, see converting PDF to Markdown.

Convert a whole folder

Click Add folder and MD Factory converts every supported document in it. Each document gets its own Markdown file next to it, which suits the per-file limits: ChatGPT's 80 files every 3 hours, Claude's 20 files per chat and NotebookLM's 50 sources.

MD Factory does not join documents. If a folder holds more files than the tool allows, join the Markdown files yourself and upload one file instead. Our guide to combining documents into one Markdown file shows the PowerShell command and a source line for each document. The batch conversion guide covers folders in more detail.

Files that are not on a tool's list can also be converted. Outlook .msg and .eml emails and ZIP archives, for example, are not on any of the three lists, and MD Factory converts them.

What conversion does not fix

  • Platform caps stay. Conversion changes what you send. It does not raise a limit. If the text of a document is longer than 2 million tokens or 500,000 words, split it.
  • Visuals are lost. Markdown holds text and image files. If your question is about a chart, send the PDF, within Claude's 100-page limit for visuals. See PDF, Word, or Markdown for AI.
  • Very large collections. When you have hundreds of documents, uploading each to a chat is the wrong tool. See PDF to Markdown for RAG.
  • Conversion can make mistakes. OCR and tables need a check against the original, especially numbers and names.

Sources

Checked on 27 October 2026.