Can Claude Read Images? Yes, Including Diagrams and Charts Inside PDFs
Every current Claude model accepts images as input. It reads text in screenshots and photos, explains charts and diagrams, and compares several images at once. Inside a PDF it sees the pictures too, because Anthropic converts each page into an image alongside its text.
What it won't do is name people in photos, count large crowds of objects exactly, or create a new image.
Facts from Anthropic's vision and PDF documentation · Independent site, not affiliated with Anthropic
What Claude Can Read From an Image
All current Claude models, from Haiku 5.5 to Fable 5.1, take text and image input. These are the kinds of images it handles well.
Text in Screenshots and Photos
Printed text, signs, receipts and legible handwriting. Ask for a transcription, a translation or a summary.
Charts and Graphs
Axes, trends and values from bar, line and pie charts. Anthropic lists analyzing charts in financial reports as a core use case.
Diagrams and Flowcharts
System diagrams, process flows and annotated drawings, explained step by step.
Tables and Forms
Rows and fields pulled out of a photo or scan and returned as structured text or JSON.
App and Website Screenshots
Layout, labels and errors in a UI. Claude Code uses this to debug a page from a screenshot.
Several Images at Once
Up to 20 per message in claude.ai. Label them Image 1, Image 2 and Claude can compare them and refer back later.
Can Claude Comprehend a Diagram or Image in a PDF?
Yes. Anthropic's PDF documentation says each page is converted into an image and its text is extracted alongside it, so charts, diagrams and photos on the page are part of what Claude reads.
Every Page Is Also a Picture
Claude gets both the extracted text and an image of each page, so a chart with no text layer is still visible to it.
Scanned Pages Work Too
Because each page is processed as an image, a scanned page without selectable text can still be read from the picture.
Cost per Page
Text typically costs 1,500 to 3,000 tokens per page, plus the image cost of the page itself.
Page Limits
The API takes up to 600 pages per request, or 100 on models with a smaller context window. The Help Center lists 1,000 pages per PDF in claude.ai.
Dense PDFs Fill Up First
Many small-font pages, complex tables or heavy graphics can fill the context window before the page limit. Split them into sections.
No Password-Protected Files
Encrypted or password-protected PDFs are not supported. Remove the password before uploading.
How to Make PDF Images Readable for Claude
Most reading errors come from the image, not the model. These steps follow Anthropic's image guidance.
Export the Page That Matters
If one diagram carries the answer, export that page or crop the diagram as a PNG and send it on its own.
Keep Text Legible
Anthropic warns that very small images, under about 200 pixels, are where Claude makes mistakes. Small labels need a larger image.
Don't Oversize Either
Claude 4.7 and later models downscale anything over 2576 px on the long edge; older models stop at 1568 px. Bigger files only add upload time.
Go Easy on Compression
Heavy JPEG compression blurs small text. Use PNG or high-quality JPEG for screenshots and charts.
Rotate It Upright
Rotated or sideways images are a listed accuracy risk. Straighten scans before uploading.
Image First, Question After
Anthropic notes Claude works best when images come before the text that asks about them.
What Claude Can't or Won't Do With Images
From the limitations section of Anthropic's vision documentation.
Name People
Claude won't identify who someone is from a photo.
Count Precisely
Counts of many small objects are approximate.
Exact Positions
Coordinates and spatial answers are approximate and worth checking.
Spot AI-Generated Images
Claude can't reliably tell whether an image was made by AI.
Read Diagnostic Scans
It isn't designed to interpret CT or MRI scans; treat answers as general information only.
Create or Edit Images
Claude only reads images. Generating or editing one needs a connected image model.
How to Give Claude an Image
Pick the surface you already use.
claude.ai and the Apps
Upload or drag images into the message box: up to 20 per message, 10 MB and 8000 × 8000 px each.
Claude Code
Paste a screenshot with Ctrl+V, or give Claude the file path and it opens the image from your disk.
The API
Send an image block as base64, a URL or a Files API file_id: up to 100 or 600 images per request depending on the model.
Ask a Specific Question
"What are the three values on the y-axis?" gets a better answer than "What do you see?"
Can Claude Read Images? FAQ
Have another question? Email us at support@claudeimagine.com.
Facts from Anthropic's vision documentation, PDF support documentation and the Help Center article Upload files to Claude, checked October 9, 2026. How many images fit in a message or chat: Claude image upload limit.
Claude Reads Images. Give It One That Makes Them Too.
Connect Claude Imagine and Claude can generate and edit images in the same chat where it reads yours. A new account starts with 9 credits.