Vision Model
An AI model that can analyze images.
Text models that learned to see.
Beyond text
A text-only model reads words. A vision model reads words and images. You can show it a photo, ask a question about it, and it answers. It uses the same magic as a text model, and it was trained on billions of image-text pairs.
What they can do
A vision model can read text in images with OCR, describe what is happening, and spot objects and faces. It can analyze charts and diagrams, and answer questions about visual content. All of this was learned from examples during training.
Hand the model an image
The picture
A crumpled coffee-shop receipt
What the model reports back
- read textReads "Blue Bottle Coffee" at the top
- read textReads the line item "Latte $5.50"
- extracted dataPulls the total: $12.75
- extracted dataFinds the date: 2026-03-14
Same model, different image, same shape of answer: it reads any text, names the objects, and pulls the numbers. All of that was learned from examples during training. None of it was hand-coded for this picture.
The models that see
Claude and GPT-5.6 have vision, and so do Gemini and Grok. Vision is now standard in flagships. You usually pay a bit more per query when you include an image.
For heavy image work (scanning thousands of documents), specialized vision models exist, but flagship LLMs cover most use cases.