How long can a single ChatGPT message be? The 10,000 character rule

The honest answer is that there is no published number

Worth stating plainly, because plenty of pages confidently quote one.

OpenAI documents its file limits precisely: 512 MB per file, 2 million tokens per text or document file, about 50 MB for spreadsheets, 20 MB per image.

There is no equivalent published figure for message length. Any specific character count you find is somebody's observation presented as documentation, and observations of this kind go stale quickly.

What actually constrains you

Three different things, and they behave differently.

The context window. How much of the conversation the model holds at once. Exceed it over a long chat and earlier messages fall out of view, which is why a long conversation starts forgetting what you established at the start. It is not a per-message limit, it is a running total.

Your browser. Pasting an enormous block of text into a web input can be truncated by the page itself before ChatGPT sees any of it. This is the most common cause of "my long paste got cut off", and it happens silently.

The file token cap, if you upload rather than paste. 2 million tokens per file, which is a documented limit and a genuinely hard one.

The one number that does exist: 10,000 characters

Not a limit on message length exactly, but it is the closest thing OpenAI publishes and it changes what a long message costs you.

Paste more than 10,000 characters into the composer and ChatGPT stops treating it as text. The content becomes an attachment instead, which is why a long paste sometimes appears as a file card rather than in the message box. The threshold was raised from 5,000 characters in June 2026.

Pasting long content sensibly

1. Break it into parts under 10,000 characters

   Small enough to stay a message rather than becoming an attachment, large enough that you are not sending thirty of them.

2. Say what is coming, before any content

   "I am pasting a long document in several parts. Do not answer until I say the last has arrived. Reply only 'received' after each."

3. Label each part

   "Part 3 of 7." It keeps the order clear and makes a missing part obvious.

4. Check it all arrived

   Ask what the final part said. If it cannot answer, something was truncated or skipped.

That last step matters more with pasting than with uploading, because browser truncation gives you no error at all. The paste looks like it worked.

The useful consequence

Since messages are not uploads, a paste under 10,000 characters does not consume any upload allowance.

Not the Free plan's 3 file uploads per day, not the rolling 80 files every 3 hours, and not the 25 GB storage cap. None of them count messages.

That makes pasting in parts the reliable route when you are capped on uploads and need to keep working, which is a more practically useful fact than any character count would have been. The one thing that breaks it is pasting too much at once, because that is no longer a message.

When to upload instead

Structure matters. Headings and tables survive an upload more cleanly than a paste.

It is a spreadsheet. Spreadsheets are analysed as data and are exempt from the token cap.

It is genuinely long. Uploading handles a long document in one action, up to the point where it gets truncated. Pasting the same content means managing the parts yourself.

It is a scan. Neither route works. You need OCR before there is any text to send at all.