Base64 Tensors & Calculations — Vision & Multimodality

[ ABORT TO HUD ]
SEQ. 1

Base64 Tensors & Calculations

👁️ Vision & Multimodality⏱ 15 min⭐ 650 BASE XP⌨ HANDS-ON LAB

Direct Optical Processing

Claude treats images and PDFs as first-class citizens in the messages array via structured content blocks:

  • Base64, URL & File Images: Provide a source object with type: "base64" (image/jpeg, image/png, image/gif, image/webp), type: "url" (direct HTTPS URL), or type: "file" (Files API file_id).
  • PDF Documents: Provide a document block with type: "base64", type: "url", or type: "file" (application/pdf).

Resolution and Token Costs

Claude maps images into visual grid tokens (~ceil(width/28) * ceil(height/28)) and automatically scales oversized inputs (1568px standard long-edge optimal cap, up to 2576px on high-resolution tiers). A typical 1024x768 image costs ~1,100–1,600 input tokens.

PDF Document Processing

Claude natively ingests multi-page PDFs both visually (charts, tables, diagrams) and textually—ideal for contract review, engineering blueprints, and financial filings.

⌨ HANDS-ON LABSubmit Multimodal Base64 Image Tensors
⭐ +200 XP

Send image data to Claude's vision encoder. Package a base64-encoded image into the user content array alongside a text prompt.

1POST to /v1/messages passing a content array with both an image block (type: image, source: base64) and a text prompt block.
lab-sandbox — simulated environment
INFINITY LAB SANDBOX v2.6 — simulated shell
Type the command for the current objective. Helpers: "hint", "solution", "clear".
$
OBJECTIVE 1 / 1 — type "hint" if stuck
SYNAPSE VERIFICATION
QUERY 1 // 2
How are visual image payloads passed to Claude in the Messages API?
Email attachments
Inside content blocks with type 'image' and a structured source object (base64, url, or file_id)
Only via external FTP
Plain text descriptions only