Vision (modèles multimodaux)
An AI model's ability to read an image the way it reads text: a screenshot, a document photo, a chart, a mockup. You send the image in the same call as your question, and the model describes, extracts, compares or reasons over it. It's what moves an assistant from 'understands text' to 'understands what I show it'.
Strengths
- Removes the need for a separate OCR pipeline for most documents
- Reasons over visual content, not just text extraction (understands a chart)
Limitations
- Fragile accuracy on dense numbers and small print
- Higher per-image cost than a plain text call
Best for
- Extracting data from scanned documents or screenshots without coding an OCR
- Giving visual context to an agent (interface, mockup, diagram)