What file types can ChatGPT read in 2026? 7 fail silently

The 7 uploads that go through and still fail

Every row here is an upload that appears to work. The file attaches, the spinner finishes, no error appears, and then ChatGPT either says it cannot find anything or answers from a fraction of what you sent.

Two patterns run through the list. Five of the seven fail because ChatGPT could not find text inside a file that looks full of it. The other two fail because the file was fine and there was simply more of it than one upload carries.

Formats that fail loudly are a smaller problem, because at least you know. MP4, MOV, MP3 and WAV are not among the four families OpenAI lists. PSD, DWG and HWP are not common document formats. A password protected PDF cannot be opened at all and usually produces something went wrong analysing the file. In every one of those cases the answer is to convert before uploading, not to retry.

Why most of them fail: the text is kept and everything else is thrown away

This single rule accounts for more confused uploads than every unsupported extension put together, and it is documented rather than folklore. From OpenAI's File Uploads FAQ, on images inside documents:

All other plans and document files only support text-based retrieval. This means that ChatGPT will extract digital text from the file and discard any images.

Visual retrieval for PDFs exists on ChatGPT Enterprise and nowhere else. On Free, Go, Plus, Pro and Business, an upload is reduced to its digital text and everything visual is dropped on the way in.

The same rule explains a chart in a quarterly report going unmentioned, a diagram in a slide deck being ignored, and a Word document that is mostly screenshots producing a thin, vague summary. None of those are failures of understanding. The images never arrived.

What OpenAI actually says is supported

Asked directly which types are supported, the File Uploads FAQ answers in one sentence:

All common file extensions for text files, spreadsheets, presentations, and documents.

That is the whole answer. Four families, no extensions, no table. Any article handing you a definitive list of forty extensions either made it up or inferred it from testing, which is worth something but is not the same thing.

Within those four families, this is what holds up consistently.

Code files are worth a note of their own. There is no special handling for them, which is good news: a .py, a .rs, a .sql or something obscure is just text as far as the upload is concerned, and it goes through fine. The problems with large code files are about length, never about the extension.

Video and audio are not in the four families

Neither video nor audio appears in the list OpenAI publishes, which names text files, spreadsheets, presentations and documents.

This trips people up because ChatGPT clearly does handle speech. That is voice mode, a different feature, and it is not the same thing as attaching an mp3 to a conversation for analysis.

If you have a recording you need worked on, get a transcript first. Any transcription tool produces a text file, and a text file is the most reliably supported thing there is. For video with subtitles already attached, an .srt or .vtt file is quicker still.

How to tell whether your own format will work

For anything unusual, this takes about thirty seconds and settles the question without a wasted upload.

1. Open the file in a plain text editor

   Notepad, TextEdit in plain text mode, or anything similar. Do not open it in the app that made it.

2. Look at what appears

   If you can see recognisable words and sentences, ChatGPT can extract them too. If you see binary noise, or a few readable fragments floating in it, there is no text layer worth reading.

3. If it is readable, upload it as it is

   The extension almost never matters at this point. A readable file in an odd format goes through the same way a .txt does.

4. If it is not readable, export rather than rename

   PDF or DOCX for anything document shaped, TXT or CSV for anything that is really text or a table, a transcript for audio and video. Changing the extension by hand is the one move that makes things worse, because it pushes an unreadable file past the picker and into a parser that cannot handle it.

5. If it was readable and the answer is still thin, the format was never the problem

   That is a length problem, and the next section covers it.

Being a supported type is not the same as being an acceptable size

Four separate ceilings apply on top of the format question.

The token cap is the one that catches people, and it is the seventh row of the failure table above. A long document can be a few megabytes, nowhere near the size ceiling, and still hold more text than one upload carries. When that happens there is usually no error at all. The file uploads, and ChatGPT answers from the part of it that fitted.

If that is what is happening to you, the format is not your problem and there are several ways to bypass the ChatGPT file upload limit using a document it already accepts.

How many files you can send, and how often

Separate again from types and sizes, there are caps on volume.

• Free accounts: 3 file uploads per day.
• Generally: 80 files every 3 hours. Rolling, so it does not reset at midnight, and failed attempts can count against it.
• Per message: 20 files per message on web. This doubled from 10 on 13 February 2026, so any page still telling you 10 was written before that. OpenAI did not say whether the mobile apps match.
• Custom GPTs: 10 files per GPT, for the lifetime of that GPT rather than per chat. Largely academic now, since GPT creation has been removed from personal accounts and survives only in Business, Enterprise and Edu workspaces.
• Projects: 5 files per project on Free, 25 files per project on Go and Plus, 40 files per project on Pro, Business, Education and Enterprise. The File Uploads FAQ still prints 20 for Plus and still calls Business "Team", so OpenAI's two pages disagree with each other. The Projects article is the more specific of the two.

Storage is a third kind of cap, and the reason people hit an upload limit having uploaded almost nothing recently. There is 25 GB per user and 100 GB per organisation, shared across chats, Projects and custom GPT knowledge. Since May 2026 there is also a much smaller Library allowance, 500 MB on Free, which is the one that actually bites first. Deleting old chats no longer clears it, because a deleted chat does not delete the files it contained. Settings, then Storage, is where that is managed now.

When the format is right and the document is simply too long

The honest manual answer is above and it works: convert to text, cut the document into parts, paste them in order with an instruction to hold the answer until the last one arrives. Keep each part under 10,000 characters, because past that point ChatGPT converts the paste into an attachment by itself and spends an upload from the allowance you were trying to protect.