Claude Certified Architect – Foundations Practice Question 96: Multiple Images Labeling
Claude Certified Architect, Foundations · Test 10 · question 6 of 10
An insurance claims tool sends Claude three photos of vehicle damage (front, side, rear) and asks "Which photo shows the most severe damage, and why?" The current request puts the question first, then the three image blocks with no text between them. Answers sometimes refer to the wrong photo. Which restructuring follows Anthropic's vision guidance?
Show answer and explanation
Answer:
A is correct. The vision documentation says Claude works best when images come before text (images placed after text or interleaved still perform well, but an image-then-text structure is preferred when your use case allows it). It also advises that when you send several images, you introduce each with a short text label (Image 1:, Image 2:, and so on) so the prompt and follow-up turns can refer to them by name.
B is wrong because the guidance is the opposite: put the images first and the question after. C is wrong because a request can carry many images (up to 100 per request on models with a 200k-token window, and 600 on others), analyzed jointly, and merging photos loses resolution. D is wrong because images belong in image content blocks of a user message; the system prompt is for instructions and role, and separating them offers no benefit.
Reference: Vision - Claude API Docs