top of page
Search

How Claude Reads PDF Files and What Its Capabilities Are in 2025

ree

Claude can read PDF files both as text and as visual documents, meaning it is able to interpret not only written content but also tables, graphs, and layout elements. The feature is available in the Claude web and app interface, in Projects, and via API integrations such as Anthropic’s own developer API, Amazon Bedrock, and Google Vertex AI. This makes PDF handling one of the most practical functions offered to users working with complex documents.


What Claude can do with a PDF.

Claude can extract and summarize text, identify tables and charts, translate sections, generate structured outlines, and produce datasets based on tabular data. It can also cite passages, report page numbers, and reformat content into structured outputs such as JSON or into new file types like Excel or Word. In 2025, Claude has expanded beyond reading PDFs to include creating and editing PDF documents directly in chat or desktop app environments, making it possible to generate new reports or convert PDFs into presentations and spreadsheets.


The limits and requirements for PDF analysis.

The limits depend on whether the PDF is processed in the interface or via the API, and whether analysis is performed visually or as plain text.


Limits in the user interface:

  • Maximum 30 MB per file.

  • Up to 20 files per chat.

  • In Projects, “unlimited” files may be added to knowledge bases, provided they remain within the context window.

  • Visual analysis, which includes text plus charts and layout, is available for PDFs of 100 pages or fewer when using Claude 4, Claude 3.7 Sonnet, or Claude 3.5 Sonnet. Beyond 100 pages, analysis falls back to text-only mode.

  • Files must be standard PDFs, not password-protected.


Limits via API:

  • Maximum 32 MB per request.

  • Up to 100 pages for visual analysis.

  • PDF must be standard and free of encryption or password protection.

Environment

Maximum size

Pages for visual analysis

Notes

Chat/Projects (UI)

30 MB per file

≤ 100 pages

20 files per chat; unlimited in Projects subject to context window

API (Messages/Files)

32 MB per request

≤ 100 pages

Standard PDFs only; three ways to send: URL, Base64, or Files API


Which models support PDF reading.

Visual PDF analysis is supported by Anthropic’s most advanced models. This includes Opus 4.1, Opus 4, Sonnet 4, and Sonnet 3.7. Earlier versions such as Claude 3.5 Sonnet and Claude 3.5 Haiku are now deprecated but still recognized in documentation. In Vertex AI and Bedrock, the same models are listed as compatible for vision-based PDF processing.

Platform

Models with PDF vision support

Notes

Anthropic API

Opus 4.1, Opus 4, Sonnet 4, Sonnet 3.7

Maximum 32 MB, ≤ 100 pages for vision

Google Vertex AI

Same set of models

Inherits the same page and size restrictions

Claude web/app (chat and Projects)

Claude 4, Sonnet 3.7, 3.5 Sonnet for vision

30 MB per file, 20 files per chat, vision limited to 100 pages


How PDF uploads work in Claude.

In chat, a PDF can be added as an attachment or dragged directly into the conversation. In Projects, multiple PDFs can be stored together in a knowledge base for reference across conversations. For reliable results, prompts should refer to specific page ranges such as “Summarize pages 14 to 22 and extract all numeric values in the tables.” This ensures that Claude grounds its answers in identifiable sections.


How PDF uploads work via API.

Anthropic supports three official methods of PDF submission:

  1. Sending a URL to the PDF.

  2. Encoding the PDF as Base64 in a document block.

  3. Using the Files API, uploading once and then referencing the file by file_id.

The Files API is recommended for repeated use, as it avoids resending large files. The developer documentation provides examples in Python, TypeScript, and curl. When using PDF vision, the file_id or equivalent reference must be paired with a clear instruction in the Messages API about what to extract and in which format.


PDF analysis in Amazon Bedrock.

On Amazon Bedrock, Claude’s PDF handling is divided into two modes:

  • Converse Document Chat: text-only extraction, excluding images, diagrams, and formatting.

  • Claude PDF Chat: full visual analysis, including images and layout, available only when citations are enabled in the API. This mode consumes more tokens but yields higher fidelity in reproducing document structure.

Mode

What it reads

Requirement

Approximate token use (per 3 pages)

Converse Document Chat

Text only

No citations required

~1,000 tokens

Claude PDF Chat

Text, images, diagrams, layout

Citations required

~7,000 tokens


The new ability to create and edit PDF files.

As of late 2025, Claude has introduced the ability to generate and edit PDFs, Word documents, Excel files, and PowerPoint slides. The feature is enabled under Settings → Features → Experimental → Upgraded file creation and analysis and is currently available to Max, Team, and Enterprise plans, with rollout to Pro expected. This gives Claude both input and output control over documents: it can read a PDF, extract structured information, and then produce a new PDF or presentation containing reformatted results.

Function

Where enabled

Plans

Purpose

Reading/analyzing PDF

Default (UI and API)

All plans with compatible models

Summarization, Q&A, table extraction, translation

Creating/editing PDF

Settings → Features → Experimental

Max, Team, Enterprise (Pro in rollout)

Generate new PDFs, convert PDFs into reports, slides, or spreadsheets


Practical guidance for using PDFs with Claude.

  • Split long files: If a PDF is over 100 pages, break it into smaller parts to keep visual analysis active.

  • Reference page numbers: Always specify which pages to analyze for more accurate, grounded responses.

  • Choose the right model: For data-heavy PDFs, use Opus 4 or Sonnet 4, and in Bedrock enable citations for full PDF vision.

  • Use the Files API: For repeated analyses of the same documents, store the file once and reference it by ID.

  • Test vision mode: Ask Claude to describe graphs or page layout to confirm whether it is in visual or text-only mode.

Claude’s PDF handling has matured into a flexible system: from simple extraction and summaries, to vision-based analysis of charts and diagrams, and now to file creation and editing. This makes it a valuable tool for users managing large and complex document workflows in 2025.


_________

FOLLOW US FOR MORE.


DATA STUDIOS

bottom of page