PDF support via API
PDF support via the Claude API allows developers to send PDF documents directly to Claude for analysis, extraction, and understanding. Rather than requiring pre-processing or plain-text extraction on the developer's side, Claude treats each PDF page as a hybrid of visual and textual content — processing the page layout as an image while also reading the underlying text layer. This means Claude can handle complex multi-column layouts, embedded charts, scanned pages, and data tables with spatial awareness that a simple text-strip approach would miss.
Developers can transmit PDFs in three ways: base64-encoded inline within a request, as a publicly accessible URL, or by referencing a file ID generated through the Anthropic Files API. The Files API approach is recommended for production use because it lets you upload a file once and reuse it across multiple API calls without re-uploading the full payload each time.
PDF support is subject to the same limitations as Claude's vision capabilities. Each page consumes roughly 1,500–3,000 input tokens depending on content density, and requests are capped by both page count and total payload size. The feature is available to any developer with API access and works across all active Claude models.
When you’d use it
- ◆Executive Document Summarization — A product team uploads quarterly earnings PDFs and asks Claude to produce bullet-point summaries of revenue, expenses, and forward guidance — including numbers from charts that a plain-text extractor would miss.
- ◆Legal Contract Interrogation — A procurement application uploads a 60-page vendor contract and queries specific clauses such as auto-renewal terms, liability caps, and payment schedules, avoiding the need to build a custom RAG pipeline for mid-sized documents.
- ◆Invoice and Receipt Data Extraction — An accounts-payable system processes batches of scanned invoices, extracting vendor name, invoice number, line items, and totals as structured JSON for direct database insertion.
- ◆Academic Paper Analysis — A research tool ingests multi-column scientific papers and extracts methodology, key findings, and citations — correctly interpreting layout even when standard PDF text-extraction would scramble column order.
- ◆Financial Chart and Table Extraction — A fintech application sends SEC filings or annual reports to Claude, which reads embedded graphical tables and charts visually, outputting structured CSV or JSON data for downstream analysis.
What changed recently
- ◆2025-04 — Files API launched in public beta (header: anthropic-beta: files-api-2025-04-14). Enables uploading PDFs once and referencing them by file_id across multiple requests, eliminating repeated base64 encoding.
- ◆2025 — Extended context models (1M-token window) increase the per-request PDF page limit from 100 pages to 600 pages.
- ◆2025 — Claude Opus 4.7 introduced high-resolution image support with a maximum image resolution of 2,576 pixels on the long edge (up from 1,568 pixels), improving analysis of dense charts and fine-print documents.
This is the short version
The full chapter has three worked examples, the common pitfalls, and the workflow that makes it pay — plus the other 84 features, kept current.
Get Claude Master — $97 →