How many knowledge files can a custom GPT hold? 10, for its lifetime

Lifetime, not per session

OpenAI's wording is specific: up to 10 files per GPT for the lifetime of that GPT.

That catches people because most other limits here are rates. The daily cap refreshes daily, the rolling cap refreshes over three hours. This one does not refresh at all.

Making ten files carry twenty documents

The ceiling counts files, not content. So the answer to needing more is almost always to combine.

Group by subject. All your pricing documents in one file, all your policies in another. Ten subject areas is a lot of coverage, and it maps better to how the GPT will be asked questions.

Use clear headers. When concatenating, put an obvious marker before each original document:

=== 2026 pricing, updated March ===
...

=== Refund policy ===
...

Without those the GPT cannot tell where one document ends and the next begins, and it will blur them together.

Convert to plain text first. Smaller, cleaner, and it strips the images that would have been discarded anyway. It also lets far more content fit under the size limits.

Leave out what will never be asked about. A tighter knowledge base produces better answers than a comprehensive one, because there is less irrelevant material to sift.

The other limits that still apply

Ten files is the count. Each of them is still subject to everything else:

That last row deserves attention. GPT knowledge files count against the same 25 GB pool as your chats and Projects. A few GPTs each holding large reference documents adds up quietly, and it is a common reason for hitting "upload limit reached" while feeling like you have uploaded nothing.

Images in knowledge files

The same rule as everywhere else. Except on Enterprise, ChatGPT extracts digital text and discards images.

So a knowledge file that is a scanned handbook contributes nothing at all. A product sheet whose specifications live in an image contributes nothing. The GPT will answer around those gaps rather than saying they exist.

If a document matters and is visual, transcribe the important parts into text before uploading it.