Asked about Images and documents as input

How do you control the cost of image input?

An answer that works

Resize before sending, since most documents are readable well below original resolution and image tokens are expensive. Send one page per call rather than batching many, which also improves accuracy.

Reading an answer is not being able to give one. The section that teaches this is Images and documents as input, 5 hours of Multimodal input.

Learn this properly

Also asked about Multimodal input

When would you use a multimodal model instead of OCR?What are the failure modes of image extraction?How does an image become something a language model can read?Design a voice assistant for a restaurant's phone line.