Pass files as base64, binary file object or URL in prompt with URL context understanding.

File size limits

• URL in prompt: 80 MB
• Base64: 20 MB
• Binary file object: 20 MB

URL

Pass a publicly accessible file URL directly in the prompt text. The model fetches and reads the file at inference time.

This is great for handling large files that don't fit in the context window.

Base64

Follows the same pattern as OpenAI file handling SDK.

Read a local file, encode it as base64, and pass it in the message content using the file content type.

Binary File Object

Read a file as a binary Buffer or Blob (TypeScript) / bytes (Python) and pass it directly in the message. The SDK handles serialization automatically.