PRODUCTION SCENARIO
A building inspector uploads an engineering PDF containing diagrams, tables, and handwritten annotations. The application must answer questions that depend on both the visual layout and the text.
How should the PDF reach the model?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: As a document block whose source is the PDF file, sent with the question
In a document block the system converts each page into an image and extracts that page's text alongside it, so Claude reasons over both and can answer about the diagrams and the handwriting. A text-extraction library discards everything that is not a character. Page images preserve the layout but drop the text layer the document block would have supplied for free. container_upload sends the file to the code execution tool for programmatic parsing, which is a data-analysis path, not visual question answering.
Official Source: https://platform.claude.com/docs/en/build-with-claude/pdf-support ↗