Too many screenshots for ChatGPT or Claude? Send their text as one file.
Claude takes 20 images a message, and ChatGPT Free 3 uploads a day. Read here, a whole folder of screenshots is one file.
Drop a folder or files
One file comes back, usually in under a second.
Paste a public link
Read at the vendor docs, October 2026. No sign-up. See what we measured
How many screenshots can you send?
Twenty a message to Claude, and to ChatGPT whatever is left of the file quota. Both count each screenshot as one, however little it shows.
| Where | What it caps | Limit today |
|---|---|---|
| Claude, a message | Images you attach | 20 on claude.ai |
| Claude, an image | Dimensions | 8000 x 8000 pixels |
| ChatGPT, every plan | Files you upload | 80 every 3 hours |
| ChatGPT, Free plan | Files you upload per day | 3 |
| ChatGPT, an image | Size | 20 MB |
Figures as of October 2026, from Anthropic's vision docs and OpenAI's uploads page, which adds that OpenAI may lower its limits during peak hours.
A screenshot's text is smaller than the picture.
We read 9 screenshots of real pages. Their text came to 631 tokens at the median. Anthropic counts the same picture at 1,560 to 2,691 tokens, by model.
| Screenshot of | As text | Words on the page |
|---|---|---|
| Wikipedia, Large language model | 800 | 97% |
| Wikipedia, Optical character recognition | 794 | 98% |
| MDN, Array.prototype.map() | 339 | 72% |
| Hacker News, front page | 1,074 | 81% |
| GitHub, facebook/react | 631 | 80% |
| Python tutorial, introduction | 545 | 99% |
| RFC 9110, HTTP semantics | 763 | 98% |
| Effective Go | 288 | 99% |
| The Rust book, getting started | 338 | 100% |
Measured 2026-10-26: the top of each page at 1920 x 1080, read with the English reader FileConcat uses and counted with its token estimate. "Words on the page" is the share of words read back that appear in the page's own text. A picture's cost is Anthropic's formula for a 1920 x 1080 image, from its vision docs, not a count of ours. OpenAI's chat app publishes no image cost.
When to send the picture instead.
Writing only
Reading keeps the words in each screenshot, under its file name. Layout, colors and anything drawn are left behind.
Screenshots of text read well. On the article and docs pages above, almost every word read back was on the page. Busy interfaces read worse: on GitHub, Hacker News and MDN, about one word in five was a misreading of a menu, an icon or a badge.
So send the text when the words are what you want answered: an error message, a long thread, notes, a page of a document. Send the image when the picture is the question, like a chart, a layout or a design, and combine the rest.
From a folder of screenshots to one file.
- 1
Drop the screenshots
Drop the folder, or select them all at once. PNG, JPEG, WebP, GIF and TIFF are read.
- 2
Press Read text
Each screenshot is read in your browser, a few seconds each. The first time, the reader for your language downloads once. Your files are not uploaded to us.
- 3
Send the one file
Copy it into the chat, or set Format to Plain and attach the .txt. Every screenshot's text sits under its own file name.
Common questions.
- How many screenshots can I send to Claude?
- Twenty images per message on claude.ai, each up to 8000 x 8000 pixels, according to Anthropic's documentation in October 2026. Read into text here first, any number of screenshots go in as one file, which is one of the twenty.
- How many screenshots can I upload to ChatGPT?
- OpenAI's upload quota counts files: up to 80 every 3 hours, and 3 a day on the Free plan. One image can be up to 20 MB. Forty screenshots read into one file cost one upload. Figures as of October 2026.
- Is a screenshot's text smaller than the picture?
- In our test, yes. Nine 1920 x 1080 screenshots of articles, docs and code pages held a median of 631 tokens of text. Anthropic's own formula counts the same picture at 1,560 tokens on its standard models and 2,691 on its high-resolution ones.
- What does reading a screenshot lose?
- Everything that is not writing: layout, colors, charts, icons and the shape of an interface. On pages of prose, 97 to 100 percent of the words read back were on the page. On busy interfaces like GitHub, Hacker News and MDN it was 72 to 81 percent. If the picture is the point, send that one as an image.
- Which languages can it read?
- It reads in your browser's language when a reader exists for it, and in English otherwise, and you can pick another before it starts. The reader for a language is downloaded once, about 3 MB for English.
- Are my screenshots uploaded to a server?
- Not to us. The text is read in your browser, and there is no account to create. The only upload is the one file you send to ChatGPT or Claude.
Send every screenshot as one file.
One image at a time instead? Image to text. Sending documents, not screenshots: the Claude upload limit and the ChatGPT upload limit.