An insurance claims team wants a Foundry-hosted model to read an adjuster's typed notes together with a photo the customer uploaded of the damaged vehicle, then produce a single summary that references details visible in the photo alongside the written notes. Which type of model should they deploy for this task?
Select an answer to reveal the explanation.
Short Explanation
The task here is not just reading notes, it is looking at a photo and reading notes at the same time and connecting the two, which means the model needs eyes as well as reading comprehension. A model that only handles text is blind to the photo entirely, it can only work with whatever the adjuster happened to type, so any dent or scratch visible in the picture but not written down just vanishes from the summary. Trying to work around that by searching the photo's file name does not help either, a file name like image47.jpg tells you nothing about what is actually in the picture. And a tool built for turning spoken audio into text is solving a completely different problem, being able to accept an uploaded file does not mean it understands what is inside that file. What is needed is a model built from the ground up to take in an image and text together and reason across both at once.
Full Explanation
The correct answer is C. Producing one summary that references details visible in the photo alongside the adjuster's written notes requires a model that can accept and reason over both an image and text together in a single request, which is what a multimodal model is built to do. Option A is incorrect because a text-only model has no way to interpret the uploaded photo at all, so any damage visible only in the image, and not already transcribed into the notes, would be missed entirely, defeating the purpose of including the photo. Option B is incorrect because a file name is just a short label and carries essentially none of the visual information in the actual photo, so keyword searching it cannot substitute for the model actually analyzing the image content. Option D is incorrect because a speech-to-text model is built to transcribe spoken audio into text, not to interpret a photograph; being able to accept a file upload does not mean it can extract meaning from an image, and it would still leave the visual damage details unaddressed.