What Is the ChatGPT File Upload Limit? (2026 Guide)

You drag a file into ChatGPT, wait, and get an error. Or worse, the upload succeeds and ChatGPT tells you it cannot read the whole thing. Either way you are now guessing at a limit that OpenAI does not display anywhere in the interface until you have already hit it.

There is not one limit. There are at least five, and they apply at the same time: a size cap per file, a different cap for images, a different cap again for spreadsheets, a count of files per conversation, and a rolling rate limit on how many you can upload in a few hours. Your account tier changes some of them and not others.

This guide sets out each limit, what actually happens when you hit one, how to work around the common blockers, and how the limits differ between the free tier and the paid plans.

What Is the ChatGPT File Upload Limit?

ChatGPT accepts files up to 512MB each, up to 10 files in a single conversation. Images are capped separately at 20MB each, and spreadsheets are effectively capped around 50MB because of how they are parsed. Text documents are additionally limited to roughly 2 million tokens per file regardless of the file size.

On top of those, there is a rolling rate limit of about 80 files every 3 hours across your account, and free accounts are restricted to around 3 file uploads per day.

Projects — the folders that hold files across multiple chats — have their own separate file counts: 5 files on the free tier, 25 on Go and Plus, and 40 on Edu, Pro, Business and Enterprise.

Total storage is capped too: roughly 25GB for an individual account and 100GB for an organisation.

Wondering how your content performs in AI search? Get a free SEO audit and see where you stand.

The Limits, One at a Time

Per-File Size: 512MB

This is the hard ceiling. Any single file over 512MB will be rejected at upload regardless of type or plan.

In practice almost nobody hits this with documents. It is a real constraint for video, large archives and raw database dumps, and it is the reason people try to split exports before uploading.

Text Documents: About 2 Million Tokens

A token is roughly three-quarters of a word, so 2 million tokens is somewhere in the region of 1.5 million words. That is a very large document — several full-length books.

The important detail is that this limit is separate from the file size limit. A 5MB plain text file can exceed the token limit while sitting nowhere near 512MB, because plain text is dense. When that happens, ChatGPT will accept the file but only work with part of it.

Spreadsheets and CSVs: Around 50MB

Tabular files are parsed differently from documents, and the practical ceiling lands near 50MB rather than 512MB. A CSV of a few hundred thousand rows will usually be fine; several million rows will not.

If you are working with data that large, ChatGPT is the wrong stage of the pipeline. Aggregate first, then upload the summary — our guide to creating a dashboard with ChatGPT covers a workflow that keeps the heavy lifting where it belongs.

Images: 20MB Each

Images have their own, much lower cap. Twenty megabytes covers almost any photo from a phone or a screenshot, but it will reject high-resolution scans, large PSDs and print-resolution exports.

Image uploads also count toward your per-conversation file count and the rolling rate limit, which catches people who are feeding in a series of screenshots.

Files Per Conversation: 10

One chat holds 10 files. This is the limit people meet most often, because it is the one that stops an in-progress piece of work rather than a first upload.

When you hit it, you either start a new conversation or use a Project, which is what Projects are for.

The Rolling Rate Limit: About 80 Files Per 3 Hours

This one is account-wide rather than per-chat. Upload 80 files in an afternoon and you will be locked out of further uploads for a few hours, even in a brand new conversation.

It resets on a rolling basis rather than at a fixed time, so waiting roughly three hours from your earliest uploads restores capacity gradually rather than all at once.

Want to know if search engines can read your pages properly? Run a free SEO audit on your site.

How the Limits Differ by Plan

The per-file size limits — 512MB, 20MB for images, the token ceiling — are the same on every plan. What changes by tier is how many files you can work with and how often.

Free accounts get the tightest treatment: around 3 file uploads per day, and Projects capped at 5 files. That is enough to try the feature and not enough to work with it.

Go and Plus lift Projects to 25 files and remove the daily upload restriction, leaving the 80-files-per-3-hours rolling limit as the practical ceiling. For most individual users this is the meaningful upgrade — if you are weighing the tiers, our comparison of ChatGPT Go versus Plus covers what else separates them.

Pro, Edu, Business and Enterprise raise Projects to 40 files and give organisations a 100GB storage pool rather than the 25GB individual allowance. Enterprise also gets visual PDF retrieval, meaning ChatGPT can interpret the layout and images in a PDF rather than only the extracted text — a genuine difference for scanned documents and design-heavy reports.

Limits at this level change fairly often. Before you plan a workflow around a specific number, check OpenAI's current help documentation rather than trusting a figure from a blog post, including this one.

How to Work Around the Limits

Split Large Files Before Uploading

The simplest fix for the 512MB ceiling and the token ceiling is to break the file up. Split a long PDF by chapter, a large CSV by date range, or a video by segment, then upload the piece you actually need.

This is usually better than uploading the whole thing anyway. A focused file produces a more accurate answer than a file where the relevant section is buried in a million words of context.

Aggregate Data Instead of Uploading Raw Rows

If you are hitting the spreadsheet ceiling, you almost certainly do not need row-level data. Group, pivot or summarise in Excel, Sheets or your database first, then upload the result.

A 200-row summary answers most analytical questions better than two million raw rows, and it uploads instantly.

Compress Images Before Upload

For the 20MB image cap, exporting at screen resolution rather than print resolution usually drops the file size by an order of magnitude with no loss of anything ChatGPT can use. It reads content, not print quality.

Use Projects for Ongoing Work

If you keep hitting the 10-files-per-conversation wall, you are using chats where you should be using a Project. Files added to a Project persist across every conversation inside it, so you upload a reference document once rather than re-attaching it each time.

Convert Awkward Formats to Text

If a file is rejected or read badly, converting it to plain text or Markdown before uploading often fixes it outright, and it sidesteps the parsing quirks that affect proprietary formats.

Curious which of your pages get cited by AI tools? Claim a free SEO audit and find out.

What Actually Happens When You Hit a Limit

The failure modes are not all obvious, which is why these limits cause so much confusion.

An oversized file fails visibly. You get an upload error and nothing is added to the conversation. This is the good case, because you know immediately.

A file over the token limit is the dangerous case. It uploads successfully and ChatGPT works with part of it, often without saying clearly that it has truncated anything. If you ask about something in the second half of a very long document and get a confident but wrong answer, this is usually why.

Hitting the rate limit produces a message telling you to try again later, with no indication of how much later. Around three hours is the practical answer.

Hitting the per-conversation count blocks the attachment but leaves the chat usable, so you can continue the conversation — you just cannot add anything new to it.

Limits Worth Knowing Before You Rely on This

File uploads are not persistent storage. Files attached to a normal conversation are tied to that conversation, and if you delete the chat you lose the attachment. Projects are the durable option, not chats.

Retrieval is not the same as reading. For large documents ChatGPT searches for relevant passages rather than reading the whole file end to end. That is why it can miss a detail that is definitely in the document — the detail was never retrieved.

Accuracy degrades with file count. Ten files in one conversation is allowed, but answers drawn across many documents at once are less reliable than answers drawn from two or three.

Scanned documents are a known weak spot outside Enterprise. Without visual retrieval, a scanned PDF is only as good as the text layer inside it, and a poor scan produces a poor text layer.

How ChatGPT Compares to the Alternatives

Claude accepts fewer files per conversation but handles very long single documents well, and tends to be more reliable at working through a long file in order rather than retrieving fragments of it.

Gemini integrates directly with Google Drive, which removes the upload step entirely for anyone already working in that ecosystem, and is the easier route if your files already live there.

Perplexity is built for search rather than document analysis. It accepts uploads, but its strength is finding and citing sources, not working through your own material in depth.

NotebookLM takes the opposite approach to all of them: it is designed around a fixed set of your own documents and will only answer from those, which is exactly what you want for a research corpus and exactly wrong for general work.

Final Thoughts

The limit most people are actually looking for is the simple one: 512MB per file, 10 files per conversation. Everything else is a secondary cap that only appears once you are working at volume.

The one worth internalising is the token limit on text documents, because it fails quietly. If ChatGPT is giving you confident answers about a very long document and getting details wrong, assume truncation before you assume the model is wrong.

For anything at real scale, treat ChatGPT as the last step rather than the first. Summarise, aggregate or split upstream and upload something focused — you will get better answers and never see a limit. If you are building repeatable analysis on top of it, our guide to creating a report with ChatGPT walks through a workflow that stays inside these constraints.

Ready to see how your site performs in search? Get your free SEO audit today.