Vision
An agent capability for understanding images, screenshots, and video — letting the model reason over visual content.
Vision is what lets an agent read a Figma file, debug a UI by looking at it, OCR a receipt, or drive a browser by understanding screenshots instead of DOM trees.
In 2026, vision is standard in frontier models. The differentiation is no longer "can it see" but "can it reason about fine detail" — reading small text, counting elements, understanding charts.
Vision-driven browser use is the killer app: instead of parsing brittle DOM, the agent looks at the screenshot the way a human would and decides where to click.
Frequently asked
Can vision agents read handwriting?+
Frontier models in 2026 handle printed text and clean handwriting reliably. Messy handwriting and cursive still trip them up.